A Call for an AI Disability Standard

Notice and Validation for People Who Use AI as a Disability Tool, Under Protections That Already Exist

Working Paper, November 2026. Draft v7 – November 7, 2026

– PDF Here – DOI – https://doi.org/10.5281/zenodo.24056038

Keywords: Artificially Intelligent Disability Assistance, AIDA, AI Disability Standard, notice, cease and desist, validation, disability credential, reasonable accommodation, reasonable adjustment, Social Security disability, ABLE, parking placard, TSA Notification Card, HIPAA minimum necessary, AI detection, disclosure penalty, smart glasses, Ley 4/2026, Checkpoint-Based Governance

Companion papers: When AI Becomes Assistive Technology: Artificially Intelligent Disability Assistance (AIDA) (Puglisi, 2026b), https://doi.org/10.5281/zenodo.22908271; When AI Policy Becomes an Accessibility Barrier: ADA Exposure from Restrictions on Artificially Intelligent Disability Assistance (AIDA) (Puglisi, 2026c), https://doi.org/10.5281/zenodo.22940224

Abstract

The first paper in this series argued that general-purpose AI can function as assistive technology. It does so when a person with a disability uses it to increase, maintain, or improve a functional capability. The second asked what follows when an institution writes a rule against that use, and found six kinds of policy that punish it. This paper holds that the protections against that punishment already exist, in the Americans with Disabilities Act, Section 504, the General Data Protection Regulation, the Employment Equality Directive, and now Spain’s Ley 4/2026. What the person lacks is a way to invoke them that the next institution will recognize. The instrument it proposes is a notice in the form of a cease and desist. The letter is dated and written for a recipient who may never have thought of AI as an accommodation. It names the protection, names the policy or action that conflicts with it, states that AI is this person’s disability tool, and explains why. It then puts the recipient on notice that any further adverse action is taken knowingly. The paper also offers a first-notice variant in the same elements, for a recipient that has an accommodation channel and has not yet acted, and asks which register works. Behind the notice sits a validation layer built the way the parking placard is built. The recipient can see the person’s standing without opening a medical file.

The paper’s purpose is to remove punishment for using AI as a disability tool. That punishment takes three forms: refusal, algorithmic punishment, and the harm of being made to label the work as AI. It ties each punishment to the protection it offends. It builds the validation from disability systems already in use rather than inventing one. Those systems are the Social Security Administration’s functional definition for eligibility, SSDI receipt as the proposed automatic qualifier with other determinations named as candidates, the ABLE account’s certification with documentation kept by the person, the parking placard’s entitlement-only face and recognition rule, the TSA Notification Card’s discreet presentation, and HIPAA’s minimum necessary principle for the notice itself. The Social Security Administration supplies the gate and the methods the validation reuses. The ADA supplies the comparator, most people in the general population, and the purpose: a person who can work is not refused the tool that lets them compete. Where those systems conflict or fall silent, the paper puts the question to the disability community and the public instead of settling it.

The law is mapped actor by actor, with what triggers each duty and what the actor must know before it attaches. The notice has force today against covered employers, funded programs, and public entities, where the sender is a person the statute protects and the need is communicated. Every European controller within the Regulation’s scope is bound by the GDPR, not by the letter, in how it handles the disability data the notice carries. In Spain the notice reaches any recipient of the right to use support products without forced deactivation. Force means a duty to consider and respond under the statute’s terms, never a duty to grant the means named. Against platforms and publishers, where the second and third punishments run at scale, no federal or Union-wide law binds the recipient to honor the notice. There the notice builds a record, and in the Union a platform’s restriction decision already carries the statement-of-reasons and complaint route the Digital Services Act gives. The paper claims nothing more. Each notice served, with its dated response, is also a record of how far refusal, penalty, and compelled disclosure reach. The paper asks those who serve notice to report what comes back. AI glasses are the stress test because one device carries every strand at once. The register throughout is exploratory, and the author is not a lawyer.

Standpoint of the Paper

This paper presents governance and policy analysis informed by statutes, regulations, cases, agency documents, and institutional practice in the United States and the European Union. The legal reading is that of a retired police officer with a Master of Public Administration whose studies included constitutional law, business law, procedural law, courts, and judicial case studies. It is not legal advice, it does not predict the outcome of any dispute, and no attorney reviewed it. Every element of the proposed instrument is a proposal, and the paper claims no legal effect for the instrument itself. Where the notice has an effect, the effect is the law’s, attached to a request the law already recognizes, and Section 2.5 says so actor by actor. Pending and closed litigation is described through docket entries and pleadings, never through a reading of what a court would decide.

The author’s position, stated for the record, is that the purpose of the instrument is to remove punishment for using AI as a disability tool. That punishment arrives in three forms. The first is refusal: a rule that bars the tool, a venue that bars the device, a platform that bars the use. The second is algorithmic punishment: a detector that flags the work, a ranking system that buries it, a classifier that reads a disabled writer’s prose as a machine’s. The third is the label, a requirement to mark the work as AI-assisted. The evidence in Section 1 shows that the label carries a measured cost to how the work is received.

The paper’s method is the method of the cease and desist letter and the parking placard. A cease and desist letter creates no right. It is “sent to a person or entity describing their alleged misconduct,” and it demands “that the specified conduct stop.” It “is a non-binding cautionary letter and does not have a legal effect,” and it “is primarily sent to later be used as evidence” (Legal Information Institute, n.d.). A parking placard creates no space. The law reserves the space, and the placard is how the attendant knows the driver may use it. The notice this paper proposes does the first job, and the validation behind it does the second. The protections come from the law; the notice invokes them; the validation makes the invocation credible on arrival. Everything else in the design is open to the community.

The paper has a second purpose, and it is why the instrument is a notice rather than a petition. No one has yet measured, institution by institution, how AI as a disability tool is met. The literature the paper cites describes the problem, and the record the notice builds is meant to count it. Each notice served, with its dated response, is one record of that, and many notices are the evidence base. The Equal Employment Opportunity Commission learned the shape of workplace discrimination from the charges filed, not from the statute. This paper proposes to learn the shape of this problem the same way, and it asks those who serve notice to report what comes back on the one rubric Section 10 states, in aggregate and under the limits stated there. The rubric gives the count a denominator, and the reports are read as the experience of those who sent notices, not as the prevalence of the problem.

Scope and method. The paper covers United States federal law and the law of the European Union, with five member states, Germany, France, the Netherlands, Italy, and Spain, chosen because each issues a graded, state-held disability credential. Primary law is cited before agency guidance, and guidance before commentary, and every statute, regulation, and case is cited from a primary record where one was retrievable. The search was not systematic. It ran through English, Spanish, French, German, Dutch, and Italian official sources as the questions required, and it concluded, with the verification of every live claim, on November 7, 2026. Where sources conflict, both are carried. The author drafted the paper with AI platforms as editor and researcher under his direction. Two rounds of review across seven platforms produced the findings dated October 30 and November 7, 2026 in Section 9, and the author decided every item.

1. The Position: The Protections Exist, and the Person Has No Way to Invoke Them That the Next Institution Recognizes

The law didn’t change when the tool did. Suppose a year arrived in which wheelchairs were declared bad. They take jobs from people who could carry, they make the user idle, they hide what a person can really do. Nothing in that year would change the ramp. An institution that locked its step would still owe one, and would owe it the day someone pointed at the law. That is where AI as a disability tool stands now, and the analogy is to the function the law protects, not to any device’s legal status. The protections exist and the tool has been declared bad, and most of the institutions bypassing the protections aren’t defying the law. No one has told them it reaches this tool.

Each of the three punishments offends a protection that is already law, where that protection’s trigger is met. Section 2 states the triggers. Refusal offends the accommodation and modification duties. Those are Title I of the ADA for employers, Titles II and III for public entities and public accommodations, Section 504 for funded programs, Article 5 of Directive 2000/78 for European employers, and Article 7.6 of Spain’s general disability law, which bars forcing a person to give up or deactivate a support product. Algorithmic punishment offends the duty to decide on evidence where that duty applies. That duty is the individualized assessment the ADA requires, the primary consideration Title II gives a request for an auxiliary aid, Article 22 of the GDPR for solely automated decisions with legal or similarly significant effects, and the human review the Digital Services Act gives for the restriction decisions it covers. The label offends confidentiality where the rule makes the person print the accommodation into the work, as Notre Dame’s engineering rule does. The protections there are the ADA’s employment rule that medical information sits in a separate file, the Commission’s rule that an employer “may not disclose that an employee is receiving a reasonable accommodation,” and Article 9 of the GDPR for every controller the Regulation reaches. The first two bind employers. For a school, FERPA governs the record and Section 504 and Title II govern the accommodation, and whether a compelled statement in a dissertation breaches them has not been tested. The paper borrows the employment rule as the design model for every recipient and says so in Section 3.4. Three disclosures are in play, and the paper keeps them apart: that AI was used, which a publisher may ask of every author as method; that an accommodation was approved; and the diagnosis behind it. The label the paper objects to is the first forced to carry the second, and no rule the paper found requires the third of anyone but the issuer. Wherever a disabled writer must wear a general label to keep the accommodation, the label imposes the measured cost of Section 1.4.

Where a rule bans the tool, penalizes its use, or forces its disclosure, the person without a disability loses a convenience. The person with one loses their footing. The rule is what preserves the natural advantage. Talent and intelligence that come without effort keep their edge because the tool that would close the gap has been taken from the one person who needed it. Remove the penalty and the gap closes, and that is the point as well as the limit. The equalizing comes from removing the penalty, not from granting anyone exclusive access. Where the tool is open to all, the disabled user’s position is restored rather than raised. Anyone should be allowed to use it, because a tool that helps a disabled person compete cannot fairly be denied to a person labeled normal. That is why the scale in Section 4, not the ban, is where the hard question lives.

The bigger problem isn’t the rule on paper. It is what happens to the person who follows it. The writer who declares AI use is pushed down the feed, the student who declares it is graded down, and the employee who declares it is marked. Public opinion punishes the honest declarer, and every honest declarer punished teaches others to hide. The notice exists so that the declaration goes to one named person, in a file, under a law, instead of to a crowd. What follows in this section is the record of the punishments themselves. It is also the record of the one thing every instance has in common: the person had no document to answer with.

1.1 Each institution decides alone

The companion paper’s model clause routes every request for AI as an accommodation to “the organization’s applicable disability accommodation process” (Puglisi, 2026c, Section 16.3). That is the right destination, and it is also the problem. Every school, employer, provider, and platform runs its own process, reads its own rule, and keeps its own file. A person who uses AI to offset a disability-related barrier explains it to each of them in turn. The record of each explanation sits in a different place under a different confidentiality rule.

The institutions that have written an answer show how different the answers are. The University of Notre Dame’s Department of Mathematics bars AI-generated content in comprehensive examinations, theses, and dissertations unless one of three conditions holds. One condition is that “Generative AI use is an approved accommodation under the Americans with Disabilities Act” (University of Notre Dame Department of Mathematics, n.d.). Its Department of Electrical Engineering permits the same accommodation and then requires the dissertation to carry a sentence: “The use of AI in this dissertation is an approved accommodation under the Americans with Disabilities Act through the Notre Dame Office of Accessibility Services” (University of Notre Dame Department of Electrical Engineering, 2025). The University of Arkansas Fort Smith permits AI “as part of an approved accommodation when authorized by Disability Services, but such tools should be functionally limited to the specific accommodation intended and not available for broader use” (University of Arkansas at Fort Smith, n.d.). The Open University in the United Kingdom permits students who have shared a disability to use generative AI as a reasonable adjustment. It does not require that use to be reported unless the assessment says otherwise (The Open University, 2025). The University of Bologna lets a student request exam adaptations only after depositing a diagnosis or certification with its disability service. It allows only adaptations consistent with that certification (Università di Bologna, n.d.). The Freie Universität Berlin’s Institute of English Philology excludes AI tools from assessed work, including stylistic revision of the student’s own prose. The page carries no accommodation exception (Freie Universität Berlin, n.d.).

Four of those rules admit the accommodation. One of the four makes the student print the accommodation into the work, which is the third punishment written as policy. None of them travels with the person. A student accommodated at Notre Dame starts over at the next institution. An employee accommodated at one employer starts over at the next, with the prior letter as evidence and nothing more.

The integrity side has noticed. The International Center for Academic Integrity published, on April 27, 2026, a statement by an academic integrity lead at the Open University: “in 2026, we are actively investigating students for using text-to-speech software, spelling correction tools, and AI-assisted writing aids that are, in many cases, specifically recommended to them through their Disabled Students’ Allowance (DSA). We are, in effect, policing the very accommodations we prescribed” (Oldham, 2026).

1.2 Platforms already act at scale

The companion paper names six types of AI policy risk: Type I use prohibition, Type II process restriction, Type III disclosure requirement, Type IV distribution or opportunity penalty, Type V discipline or exclusion, and Type VI perception enforcement (Puglisi, 2026c, Section 6). The last two types now run at platform scale. LinkedIn launched a “Seems like AI slop” reporting option on July 30, 2026, and more than one million members used it within weeks. The platform reports roughly 40 percent fewer views on content it classifies as low-quality AI content. The companion paper carries that figure with its causal limit, since new classifiers launched the same day and the flag’s isolated effect is unmeasured (Rojas, 2026; Puglisi, 2026c, Section 6.1). Medium requires “any story incorporating AI assistance be clearly labeled as such” and restricts unlabeled AI-assisted text to the author’s own network. Its definitions exempt “AI outlining, or AI assisted fact, spelling, or grammar checkers,” a function-based line with no accessibility exception (Medium, n.d.). Substack’s reader-facing detector, announced July 21, 2026 for posts over 100 words, shows a reader who requests a scan a Pangram estimate of how much of the text is AI-assisted. The writer can scan a draft first, report a detection error, and, under the help page as updated August 9, 2026, disable the analysis for a post or note (Best, 2026; Substack, 2026). The default remains a scan on request by any reader. A disabled writer on any of these platforms is reached by a system that acts on how the work reads, whether or not any tool was used.

1.3 The mislabeling thread, from 2025 to 2026

The thread that runs through this paper began in the author’s own published record. In April 2025 the author published two versions of one article, one dictated and one refined with AI assistance, and set them up for an Originality.ai review. He judged the refined version “the better content and more professional” while expecting that “the AI scan would claim it has less value” (Puglisi, 2025a). In May 2025 the author argued that detectors “measure patterns, not value” and that “The more professional and rigorous the content, the more likely it is to be penalized” (Puglisi, 2025b). That post itself scored 51 percent original. In May 2026 the author scanned an article about the encyclical Magnifica Humanitas with the same tool. The earliest draft scored “Likely AI, 100% Confident.” Mechanical prose fixes brought the score to 94 percent and a structural rewrite to 93 percent. The addition of personal voice edits, verbatim memories, and a handwritten closing returned it to 100 percent. In the article’s words, “The scanner punished the author for adding more of himself” (Puglisi, 2026a). That is one article and one tool, an anecdote that motivates the question, not evidence beside the studies that follow.

The peer-reviewed record points the same way and is now contested on its own terms. Liang, Yuksekgonul, Mao, Wu, and Zou (2023) found that detectors “consistently misclassify non-native English writing samples as AI-generated” and “may unintentionally penalize writers with constrained linguistic expressions.” Chambers and Kelley (2025) scored about 60,000 Reddit posts with OpenAI’s GPT-2 detector and found that fewer than 2 percent of posts in either group were flagged. Posts from autism-focused communities drew significantly more flags, with distinguishing features that “did not clearly align with known features of AI-generated text.” A reanalysis published September 6, 2026 reports that a replication of that protocol on a different data set found a 1.79 percent false-positive rate for autistic writers against 1.82 percent for controls. It reports that the original protocol gave 1.70 percent against 1.20 percent, and that a second detector gave 0.59 percent against 0.61 percent. The reanalysis is a self-published Substack post under a pen name, not peer reviewed, and its author discloses “$80,000 in API tokens free of charge with no conditions” from the detector vendor Pangram Labs (Dispatches from the Autism Wars, 2026). Pangram’s own accuracy post of October 16, 2026 cites that work as finding “no appreciable bias” without naming the funding (Pangram Labs, 2026). Pangram is the detector behind Substack’s reader-facing scan, so the vendor disputing the finding is the vendor whose scanner any Substack reader can run on a disabled writer’s post (Substack, 2026). The paper reports both findings at full strength. Detector flag rates are specific to the detector and the population. In the autism comparisons located they sit at or below 2 percent in both groups, and in Liang and colleagues’ corpus seven detectors misclassified over half of 91 TOEFL essays by non-native writers, an average false-positive rate above 60 percent (Liang et al., 2023). Whether the difference between the autism and control groups is real is now a dispute between a peer-reviewed study and a vendor-supported replication. Turnitin itself states that highlighted text should be used “to initiate a conversation, not to draw a conclusion” (Turnitin, 2023). The AI Incident Database records Incident 849 on detection tools misidentifying neurodivergent and non-native students’ work (AI Incident Database, n.d.).

What the dispute does not change is the design point. The figures are flag rates, since no study verified human authorship for every post it scored. A flag rate of 1 or 2 percent applied at platform scale reaches thousands of people. The person reached has no document to answer with, and no protection that reaches a flag before a human acts on it.

1.4 The label has a measured cost

The third punishment is the quietest, and the evidence on it is the strongest. Schilke and Reimann (2025) ran 13 preregistered experiments and found that people who disclose AI use are trusted less. The penalty held when the disclosure said a human reviewed the work and when it said AI was used only for proofreading. Raj, Berg, and Seamans (2026) found AI-labeled creative writing devalued by 6.2 percent on average across 16 experiments. That result supersedes the same authors’ withdrawn 2023 preprint, whose smaller preregistered series had found no effect for short stories and a penalty for first-person poems. Reif, Larrick, and Soll (2025) report “a social evaluation penalty for using AI.” Kirk and Givi (2025) report an “AI-authorship effect” in which consumers who believe an emotional message was written by AI respond less favorably, through lower perceived authenticity and moral disgust. The effect is smaller when the AI only edits. In scholarly publishing, BaHammam (2025) reports that a bibliometric review of 1,998 radiology manuscripts found only 34, or 1.7 percent, acknowledging any involvement of a large language model. BaHammam names fear of bias as the reason authors stay silent. None of these studies tested disabled users or disclosure made because of an accommodation. Schilke and Reimann’s six disclosure framings covered human review, proofreading, intent, error, and transparency, and none named a disability. The paper says so. The point stands without that test. The label costs something to everyone, and a person who wears it because of a disability cannot take it off without giving up the accommodation, which is a choice others still have.

The largest publishers have answered the penalty, and their answers diverge. Springer Nature’s editor guidance adopts a Green, Amber, Red system in which Green is “Assistive AI use (low risk).” It instructs editors that “The goal is not to identify or penalise AI use” and that “The presence of an AI declaration does not indicate concern,” citing the under-disclosure finding above (Springer Nature, 2026). Elsevier’s journal-wide policy exempts “AI tools or features within specialist disability-related assistive technology, provided these are used solely for accessibility purposes,” while requiring disclosure when a tool makes “substantive changes to sentence structure or organization” (Elsevier, 2026). One of its journals, in a guide updated September 2026, carries no such exemption and requires that even grammar tools be disclosed (Journal of Oral and Maxillofacial Surgery, 2026). SAGE treats “tools to improve language, grammar, or structure” as “assistive AI tools” that “do not require disclosure,” a category defined by function rather than disability (Sage Publishing, n.d.). Taylor and Francis exempts nothing (Taylor & Francis, 2025). Staiman (2026) argues that “Style, fluency, or perceived ‘AI-ness’ of prose should be explicitly excluded from integrity judgments.” Four publishers whose policies are in force this year have therefore answered the disclosure question four different ways. The one answer that names disability reaches specialist products only. A disabled author’s position depends on which journal the author submits to.

1.5 Without a shared instrument, a person discloses again everywhere

The record of each disclosure sits where the institution keeps it, and the Department of Labor showed in 2026 what that means. A spreadsheet of employees’ reasonable accommodation requests was emailed to a personal address outside the department’s domain by someone with authorized access. It carried categories of functional limitation along with names, duty location, title, supervisor, and pay grade, and nearly 3,150 employees who submitted requests between October 2023 and June 1, 2026 were affected (Heckman, 2026a; 2026b). That file is the level of disclosure an accommodation office already holds, and it existed in one agency. A person accommodated at five institutions has five such files.

1.6 The fixed position

A system is needed, and the system is a notice with a validation behind it. The need is held firmly. The protections exist, and the person cannot invoke them in a form the next institution recognizes. An institution that punishes the use without having been told can say it did not know. The notice ends the not knowing. The design that follows is a proposal, and every point where the source systems conflict goes to the community in Section 9.

2. The Protections That Exist Today

2.1 What a notice is, and what it is not

The paper’s instrument is modeled on the cease and desist letter because that form already does the work the paper needs and claims nothing the paper cannot claim. The letter describes the conduct, names the right it offends, demands that the conduct stop, and warns of what the law allows to follow if it does not. It is kept as evidence, and it has no legal effect of its own (Legal Information Institute, n.d.). Its force is borrowed from the law it cites and from the record it creates. Before the letter, the recipient may not have known; after it, every further act is taken with knowledge. Knowledge is not liability. What the recipient does with the knowledge is judged under the law that reaches it, and the notice decides none of that.

Disability law gives that knowledge a specific job. Under Title I, the accommodation duty attaches when the employer knows of the disability and the need. The request that supplies that knowledge “may use ‘plain English’ and need not mention the ADA” (Equal Employment Opportunity Commission, 2002). Under Title II, the public entity must “give primary consideration to the requests of individuals with disabilities” in choosing auxiliary aids (28 C.F.R. § 35.160(b)(2)). It must also modify a policy “when the modifications are necessary to avoid discrimination” (28 C.F.R. § 35.130(b)(7)(i)). In the Union, Directive 2000/78 requires “appropriate measures, where needed in a particular case” (Council Directive 2000/78/EC, Article 5). Spain’s Article 7.6 bars forcing a person to deactivate a support product “especialmente … en el uso de bienes, productos y servicios puestos a disposición del público” (BOE, n.d.). None of those duties is triggered by a credential. Each attaches to a need the institution knows of or should know of. A notice is the most direct way to make the need known, and a credential alone does not.

The knowledge element answers two questions every institution asks before it builds anything. Why is there no accessible parking by the door? Because no one showed the need, and no one said who the space was for. Why is there a ramp? Because the law recognized that a locked step is a denial, and the notice does the same work for a tool. It shows the need in the person’s own words, and it names who the accommodation is for, with the validation where one exists. It is written for a recipient who has never thought of AI as an accommodation, and it teaches before it warns.

The analogy has a limit, stated once. A placard invokes an entitlement a legislature settled before the car was parked, and the space is the same for every holder. The notice invokes an accommodation claim the law settles one case at a time, by individualized assessment. The means it asks for differs with the person and the task. The notice starts that assessment with the recipient knowing; the validation lends the claim credibility; neither settles it.

Two things the notice is not. It is not a determination of the accommodation. The Commission’s guidance reserves the choice among effective accommodations to the employer, with the person’s preference given primary consideration (Equal Employment Opportunity Commission, 2002, quoting 29 C.F.R. Part 1630, Appendix, § 1630.9). The notice therefore proposes a means and opens the process; it does not close it. And it is not a shield against ordinary rules. The TSA card the paper borrows states on its face that presenting it “does not exempt me from screening” (Transportation Security Administration, n.d.). The companion paper’s model clause keeps “essential functions, security, confidentiality, academic integrity, and fundamental alteration” as legitimate requirements (Puglisi, 2026c, Section 16.3). The notice asks for the same standard by a different means, and for no action on a reading until a named human has decided on evidence.

Two choices in the design should be visible. The architecture is a cease and desist because that is the form the law already uses to make a recipient knowing and to keep evidence. The letter is titled a Notice of Protected Assistive Use so that it arrives in the accommodation channel and stays there. A letter that announces itself as a legal demand is routed to counsel. The letter also presumes the recipient did not know. That presumption is a choice with a cost: it teaches, and it is the honest reading of most institutions today. It also gives away the first step of a later claim that the recipient acted with deliberate indifference. The person who wants that step kept can strike the sentence. Section 3.5 gives a second register for the person who wants the letter to open as a request.

2.2 United States, actor by actor

The ADA defines disability as “a physical or mental impairment that substantially limits one or more major life activities,” a record of such an impairment, or “being regarded as having such an impairment.” Its list of major life activities includes “learning, reading, concentrating, thinking, communicating, and working” (42 U.S.C. § 12102). The regulations add that the threshold question of whether an impairment substantially limits a major life activity “should not demand extensive analysis” (29 C.F.R. § 1630.2(j)(1)(iii)). A person covered solely under the regarded-as prong is owed no reasonable accommodation or modification (42 U.S.C. § 12201(h)). The notice therefore rests on the first two prongs: an impairment that substantially limits a major life activity, or a record of one. That definition, not the Social Security Administration’s, is the one the notice invokes. The Administration’s definition is the one the validation borrows in Section 4.1, and the two are different instruments for different jobs. The ADA makes failure to accommodate a form of discrimination (42 U.S.C. § 12112(b)(5)(A)) and bars disability inquiries not “job-related and consistent with business necessity” (§ 12112(d)(4)(A)). It requires medical information to be kept on separate forms in separate files as a confidential record (§ 12112(d)(3)(B); 29 C.F.R. § 1630.14(c)(1)). Title I reaches employers with 15 or more employees (42 U.S.C. § 12111(5)(A)). Title II requires public entities to make reasonable modifications in policies, practices, or procedures unless the modification would fundamentally alter the service (28 C.F.R. § 35.130(b)(7)(i)). Title III carries the parallel duty for public accommodations (28 C.F.R. § 36.302). Title II also requires “appropriate auxiliary aids and services” for effective communication and defines them to include “screen reader software; magnification software; optical readers” and “real-time computer-aided transcription services.” It directs the public entity to “give primary consideration to the requests of individuals with disabilities” in choosing them (28 C.F.R. §§ 35.104, 35.160(b)), which is the nearest existing federal handle on the captioning and reading functions of the glasses in Section 5. The auxiliary aids rule reaches communication: reading, hearing, speech. AI used for memory, pacing, or composition is a modification of policy under § 35.130(b)(7)(i), not an auxiliary aid, and the paper asks for it under that rule. Section 504 bars exclusion “solely by reason of her or his disability” in any program receiving federal financial assistance (29 U.S.C. § 794(a)). The companion paper’s actor and statute matrix maps which of these reaches which institution, with the known limits. Those include the circuit split on whether an online-only platform is a place of public accommodation (Puglisi, 2026c, Section 7).

The duty attaches on a request, and the Equal Employment Opportunity Commission’s guidance sets the bar for the request low. The individual “may use ‘plain English’ and need not mention the ADA,” and the employer may ask for documentation only “when the disability and/or the need for accommodation is not obvious.” Where more than one accommodation would work, “the employer may choose among reasonable accommodations as long as the chosen accommodation is effective,” with “the preference of the individual with a disability” given “primary consideration” (Equal Employment Opportunity Commission, 2002). Those three rules shape the notice in Section 3. It is the request, it carries the three prongs and the validation where one exists so that documentation is not the first demand, and it proposes a means without pretending to settle it. The Act also bars retaliation against a person who requests an accommodation or opposes a practice it forbids (42 U.S.C. § 12203). A notice is such a request, and an adverse action that follows it is the first fact a retaliation claim records.

The closest existing template for a camera-based aid is the service-animal rule. Staff may ask only two questions, whether the animal is required because of a disability and what task it performs. They may not require documentation or a demonstration, and they may exclude the animal only for its conduct (U.S. Department of Justice, 2010; 28 C.F.R. §§ 35.136(f), 36.302(c)(6)). That rule gates on function and conduct, not on a credential. It is written for service animals and binds no one about AI, and the paper borrows its shape. Question 3 asks how a validation squares with it.

Platforms and publishers. The companion paper’s actor and statute matrix records the circuit split on whether an online-only platform is a place of public accommodation under Title III (Puglisi, 2026c, Section 7). Two further limits sit beside it. In Moody v. NetChoice, LLC, the Supreme Court addressed the First Amendment protection of a platform’s editorial choices about third-party speech. Section 230 of the Communications Decency Act shields a provider from liability as the publisher of information provided by another (Moody v. NetChoice, LLC, 2024; 47 U.S.C. § 230(c)). The Third Circuit has since held that a platform’s own algorithmic curation is first-party speech that Section 230 does not immunize (Anderson v. TikTok, Inc., 2024). That holding removes one shield and raises the other for the same conduct, and either way a notice reaches a platform’s ranking only through a record. Whatever Title III reaches, those authorities limit what any notice can reach in a platform’s ranking, labeling, or detection of content. The paper does not claim otherwise. Against a platform or a publisher, the notice does the job a cease and desist does when no statute compels the recipient. It builds the dated record that a later proceeding would need, whether a complaint under the Digital Services Act’s statement-of-reasons and complaint provisions, a Title III claim where the circuit allows one, or a discrimination filing. And it converts an unknowing policy into a knowing one.

The floor moves

Two 2026 events show why the paper’s own principle, that the legal floor is not the ethical ceiling, is a working caution rather than a slogan. The Department of Justice moved the Title II web accessibility compliance dates by a year, to April 26, 2027 and April 26, 2028, in an interim final rule of April 20, 2026. The Department of Health and Human Services moved the Section 504 web and mobile dates in step, and the National Federation of the Blind sued both departments on May 21, 2026 (Federal Register, 2026a; 2026b; Palamountain, 2026). On September 23, 2026, in Texas v. Kennedy, filed as Texas v. Becerra, a federal court in Texas entered a final judgment on the joint motion of five states (Texas, Alaska, Louisiana, Montana, and Florida) and HHS, who described themselves as “having conferred and reached an agreement to settle this case.” The judgment vacated six integration provisions of the 2024 Section 504 rule: the definition of “most integrated setting” in 45 C.F.R. § 84.10, the final sentence of § 84.22(b), § 84.43(d), § 84.68(d), § 84.76 in its entirety, and the final sentence of § 84.93(b). It left the web and mobile provisions of Subpart H untouched (Texas v. Kennedy, 2026a; 2026b). Olmstead itself stands, and the Federation had demanded dismissal of the case in a July 2026 resolution (National Federation of the Blind, 2026). A rule the disability community relied on was removed by agreement between the agency that wrote it and the states that sued it. A notice that rests on existing law inherits that risk, which is why the paper’s design rests on several systems rather than one.

Federal AI policy is silent on the person who uses AI

Five presidential instruments since January 2025 set federal AI posture, and the paper searched each of them: Executive Order 14179 of January 23, 2025; the July 2025 AI Action Plan; Executive Order 14365 of December 11, 2025; Executive Order 14409 of June 2, 2026; and Executive Order 14434 of September 29, 2026. The last renames artificial intelligence “Super Intelligence” in non-statutory executive documents (Federal Register, 2025a; 2025b; 2026c; 2026d; White House, 2025). None of the five names disability, assistive technology, or accommodation in its text, searched by string on November 6, 2026. The Action Plan’s one use of “accessibility” concerns access to federal statistical data under the Confidential Information Protection and Statistical Efficiency Act. Silence in AI policy is not absence of protection; the disability statutes are technology-neutral, which is the paper’s point. The EEOC’s 2022 technical assistance on AI and the ADA went offline in January 2025 and returned a page-not-found error when opened on November 6, 2026 (Olive, 2026). One bill names the person’s own use. H.R. 10515, the Artificial Intelligence Education and Workforce Readiness Act of 2026, introduced September 21, 2026, would require the Rehabilitation Services Administration to provide technical assistance on “accommodations for individuals with disabilities in artificial intelligence-augmented occupations.” It would require state vocational rehabilitation plans to describe how they will “enable individuals with disabilities to (i) use artificial intelligence.” It creates no credential and had not left committee as of October 27, 2026 (H.R. 10515, 2026). State AI laws in Illinois, Texas, California, and Colorado regulate AI used on people. None gives a person a right to use AI or protects a person who uses it from an institution’s rule (Puglisi, 2026c).

What binds each recipient of the notice

HIPAA itself reaches covered entities and their business associates. The Department of Health and Human Services states plainly that “The Privacy Rule does not protect your employment records, even if the information in those records is health-related” (U.S. Department of Health and Human Services, n.d.). Employers are bound instead by the ADA’s own confidentiality rule. Medical information is “collected and maintained on separate forms and in separate medical files” and “treated as a confidential medical record,” with exceptions only for supervisors who need to know work restrictions and accommodations, first aid and safety personnel, and government compliance officials (29 C.F.R. § 1630.14(c)(1); 42 U.S.C. § 12112(d)(3)(B)). The Equal Employment Opportunity Commission’s guidance adds that “An employer may not disclose that an employee is receiving a reasonable accommodation” because doing so usually reveals a disability (Equal Employment Opportunity Commission, 2002). Schools that receive funds under a program administered by the Department of Education are bound by FERPA for the education records they keep (20 U.S.C. § 1232g; 34 C.F.R. Part 99). No federal rule names platforms, so the notice asks platforms to adopt the same principles. In the European Union, one rule reaches every controller within the Regulation’s material and territorial scope. Article 9 of the General Data Protection Regulation treats data concerning health, which includes disability, as a special category (Regulation (EU) 2016/679, Article 9).

HIPAA’s minimum necessary standard binds covered entities and business associates only (45 C.F.R. § 164.502(b)). The employer rule and FERPA cover employers and schools (29 C.F.R. § 1630.14; 34 C.F.R. Part 99). For platforms, Washington’s My Health My Data Act reaches consumer health data outside employment with a private right of action. The FTC’s Health Breach Notification Rule reaches only a vendor of personal health records after a breach (Wash. Rev. Code ch. 19.373; 16 C.F.R. Part 318). HIPAA also gives the individual “a right of access to inspect and obtain a copy of protected health information” held by a covered entity (45 C.F.R. § 164.524(a)(1)). That provision supports a person carrying the record rather than the institution holding it. The United States protection for the notice is therefore a patchwork by sector. That is why Section 3 applies HIPAA’s principles by the notice’s own terms rather than by any statute’s force.

The validation mechanics in law

The Social Security Administration’s definition and its mental listings (Section 4.1), ABLE’s certification rule (26 C.F.R. § 1.529A-2), the vocational rehabilitation presumption (34 C.F.R. § 361.42), and the parking guidelines (23 C.F.R. Part 1235) are all in force. ABLE’s 2026 contribution limit is $20,000 (Internal Revenue Service, 2025). The age change to 46 is Section 124 of the Secure 2.0 Act of 2022, enacted as Division T of Public Law 117-328 (Social Security Administration, 2026g). The regulation’s text still reads 26 at the sections cited. The statute moved the age to 46 from January 1, 2026, and the Administration applies 46. The parking rule’s reciprocity clause (§ 1235.8) is the only United States rule located in this review that requires one jurisdiction to honor another’s disability credential. The paper proposes it as the model for question 7.

2.3 European Union, actor by actor

One privacy rule for every controller

Article 9 of the GDPR treats data concerning health as a special category for every controller in every sector, with Recital 35 placing disability inside it. Article 9(2)(b) supplies the employment-law basis, Article 5(1)(c) the minimisation principle, and Article 22 bars solely automated decisions with legal or similarly significant effects based on special-category data (Regulation (EU) 2016/679). Where the United States has a patchwork, the Union has one rule. The notice’s HIPAA-style principles are already law for the handling of the data by every controller the Regulation reaches. The Dutch employer who checks whether a worker is in the doelgroepregister sees a yes or a no only (UWV, 2025), which is the minimum necessary standard in a working register.

The accommodation duty and its reach

Directive 2000/78 requires employers to take “appropriate measures, where needed in a particular case” unless they impose a disproportionate burden. The Court of Justice read the duty broadly in HR Rail (Council Directive 2000/78/EC, Article 5; Puglisi, 2026c, Section 15.1). The duty concentrates in employment. The Digital Services Act gives a platform user a statement of reasons for the restriction decisions Article 17 covers, on platforms that are not micro or small enterprises, and the statement must disclose automated detection. It adds a human-supervised complaint route under Article 20 (Regulation (EU) 2022/2065, Articles 17, 19, 20). That is more than any United States platform rule provides, and it creates no right to use AI. The Act’s own “notice” is the Article 16 mechanism for illegal content, and the paper’s notice is not that.

The AI Act and the Accessibility Act

The AI Act names disability in binding text. Article 5(1)(b) prohibits AI that exploits the vulnerabilities of a group “due to their age, disability or a specific social or economic situation” where the practice materially distorts behavior and causes or is reasonably likely to cause significant harm. Article 16(l) requires high-risk providers to meet accessibility requirements from December 2, 2027 for Annex III systems and August 2, 2028 for Annex I systems after the Digital Omnibus of July 2026. Article 95(2)(e), in force since August 2, 2026, invites voluntary codes of conduct on “accessibility for persons with a disability” (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744). Every one of those provisions concerns AI used on people. Article 95 is the only in-force Union provision located that could hold a voluntary code naming a disabled person’s own use, and a voluntary code binds no one. The Article 50 Guidelines exempt from marking “AI-generated content that only transforms authentic human input through assistive technologies allowing persons with disabilities to communicate,” while marking summaries and rewriting that changes meaning. That is the cut the companion paper describes (European Commission, 2026; Puglisi, 2026c, Section 15.2). The European Accessibility Act defines assistive technology to include software used to increase, maintain, substitute, or improve functional capabilities. It lists no AI category among the products and services it covers (Directive (EU) 2019/882, Article 3(37)).

The European Disability Card

Directive (EU) 2024/2841 creates a European Disability Card that member states must issue from June 5, 2028, mutually recognised for short stays of up to three months. It excludes social security, social protection, and employment-field benefits, long-term inclusion services, and anything tied to an individual assessment, and member states remain competent to assess disability (Directive (EU) 2024/2841, Articles 2(3), 3(7), 7; European Commission, n.d.). The European Parliament’s committee report records that the directive “does not guarantee the mutual recognition of disability status” (European Parliament, 2025). The European Disability Forum reported in August 2026 that the Commission’s delegated act on the card did not include its proposal for a tactile cue to help card users find the QR code. Nor did it include a text alternative for the encoded data (European Disability Forum, 2026a). A card can be an entitlement-only credential and still be inaccessible to its holder, which Section 9 carries as question 4’s companion.

National credentials, and Spain

Each of the five member states examined issues a graded, state-held credential after public assessment. Germany’s Schwerbehindertenausweis records a Grad der Behinderung in steps of 10 with Merkzeichen letters and no diagnosis (SGB IX §§ 2, 152). France’s reconnaissance de la qualité de travailleur handicapé runs 1 to 10 years or for life, and “La déclaration de la RQTH à l’employeur n’est pas obligatoire” (Service Public, 2026). The Netherlands keeps a register employers may query for a yes or no (UWV, 2025). Italy is consolidating its assessments into one INPS determination from January 1, 2027 (INPS, 2026). Spain grades disability in five classes with a card “válida en todo el territorio del Estado” (BOE, 2022). None of these credentials says what means its holder uses.

Spain went further on October 23, 2026, when Ley 4/2026 entered into force. Its new disposición adicional decimoséptima to the general disability law provides that persons with disabilities “tendrán derecho a beneficiarse de los sistemas de inteligencia artificial, tanto de aprendizaje y de toma de decisiones automatizadas como de inteligencia artificial generativa o cualesquiera otros que surjan en el futuro, para la promoción y protección de los derechos reconocidos en la presente ley,” and that they “tienen derecho a la protección contra los efectos discriminatorios, incluida la denegación de ajustes razonables, generados por la falta de transparencia, los sesgos y la falta de supervisión humana de los sistemas de inteligencia artificial” (BOE, 2026). In working translation: persons with disabilities shall have the right to benefit from artificial intelligence systems, including generative AI and any that arise in the future, for the promotion and protection of the rights the law recognizes. They also have a right to protection against discriminatory effects, including the denial of reasonable adjustments, generated by the opacity, bias, and lack of human oversight of those systems. The same law adds a paragraph to Article 7 of the general disability law: “Las personas con discapacidad tienen derecho a utilizar los productos de apoyo y las ayudas técnicas que necesiten … En ningún caso podrán ser obligadas a prescindir de dichos productos o a desactivarlos, especialmente en sus relaciones con las administraciones públicas y en el uso de bienes, productos y servicios puestos a disposición del público. Ello se entenderá sin perjuicio de la acreditación de su necesidad para la participación en los procesos selectivos de acceso al empleo público o a la formación especializada” (BOE, n.d.). In working translation: persons with disabilities have the right to use the support products and technical aids they need, and in no case may they be required to do without them or to deactivate them, especially in dealings with public administrations and in the use of goods and services available to the public. This is without prejudice to the accreditation of their need for participation in selection processes for public employment or specialized training.

Three features of the Spanish text bear on this paper. First, the right to benefit from AI reaches the person’s own use. The right to use support products with a bar on forced deactivation answers the first punishment, refusal, on the statute’s text as a right, with its application to a general-purpose consumer device untested. Refusing a reasonable adjustment is a grave infraction with fines up to 90,000 euros (BOE, n.d., Articles 81, 83). Second, the clause on discriminatory effects reaches harms “generados por” an AI system’s properties. That covers a detector or a screening tool used on the person and does not, on its text, reach a human rule that bans the person’s own use. Third, the need-accreditation clause is a notice mechanism written into law for one setting, public selection processes. The law defines a support product, in the dependency law’s provision service rather than in the general disability law, as “todo dispositivo, instrumento, programa informático, equipamiento o sistema técnico, destinado a prevenir, compensar, aliviar o eliminar limitaciones en la autonomía o a facilitar la participación en la comunidad” (BOE, 2026, Artículo segundo, apartado Veinticuatro, Artículo 25 bis of Ley 39/2006), a definition that names software. Whether that definition governs the new Article 7.6 is a question of interpretation. How need is accredited is left to the reasonable adjustments framework due by ministerial order within a year, by October 23, 2027 (BOE, 2026, disposición final vigesimoquinta). Spain’s protections also apply “con independencia de la existencia de reconocimiento oficial de la situación de discapacidad o de su transitoriedad” (BOE, n.d., Article 64.2). The law’s defense, arbitration, and judicial measures against discrimination therefore reach the person without the card, which Section 8 carries as dissent against the validation.

The dated record. Spain is the first jurisdiction located in this review to move in law, and this series thanks it for putting the recognition into a statute. Spain wrote into statute what this paper argues the ADA and the Union’s law already protect. Article 7.6 is a legislature’s confirmation that the issue is real and that the protection should be named, not a new right the rest of the world lacks. The Government’s bill carried the AI clause from July 15, 2025, and Parliament added the support-products paragraph before the Congreso’s approval on July 14, 2026. The Cortes gave final approval on September 16, 2026, and the law was sanctioned October 1, published October 3, and entered into force October 23 (Congreso de los Diputados, 2025; 2026; Ministerio de Derechos Sociales, Consumo y Agenda 2030, 2026; BOE, 2026). The two companion papers were published on September 23 and 24, 2026, one week after final approval and before the law was sanctioned, published, or in force. Neither cites the Spanish text, and the first paper’s search record states that non-English literature was not searched systematically (Puglisi, 2026b, Appendix A). The convergence was independent. The papers supply what the law does not: a functional test for when a general-purpose program becomes a support product for a particular person, a typology of the institutional rules that punish that use, and the notice instrument this paper proposes. Spain recognized the issue in law first, and this series identified it on its own and asks what instrument the recognition now needs.

2.4 Where each side reaches further, and where neither reaches

Where neither reaches. No credential in either jurisdiction states the means its holder uses, and attaching a stated means to an existing status would be new in every system examined. Instruments that touch the person’s own use exist without a credential: Spain’s two provisions, the Article 50 Guidelines’ exemption for assistive communication, Elsevier’s exemption for specialist assistive technology, and local university exceptions such as Notre Dame’s. None is portable, none shields a general-purpose use from an institution’s rule outside its own four walls, and none states the means in a form another institution must honor. No federal or Union-wide law gives a camera-based aid a venue access right, and Spain’s bar on forced deactivation is the one national provision that could, untested. The European Disability Card excludes the employment and education domains this paper targets, and the United States has no card at all.

2.5 Where the notice has force today, and where it builds a record only

The table states, actor by actor, what the notice does on arrival. “Force” means that a statute or regulation places a duty on the recipient that the notice’s knowledge triggers. The duty is to consider and respond under the statute’s terms, which may end in the means requested or in an equally effective alternative. It is never a duty to grant the means named. “Record” means that no located law compels the recipient and the notice’s work is evidentiary.

Remedies and clocks. The notice starts no clock of its own. The clocks below belong to the law, run from the adverse act, and are the reason the record carries dates.

In Spain the general disability law adds an infraction procedure, in which refusing a reasonable adjustment is a grave infraction (BOE, n.d., Articles 81, 83). Article 64’s defence, arbitration, and judicial measures reach the person with or without official recognition (Article 64.2).

3. The Notice

The notice is the step a person can take today, before any issuer exists. It is written as a cease and desist is written, and it borrows HIPAA’s privacy principles whether or not HIPAA binds the recipient. Those principles already set accepted rules for handling health information, and the paper borrows them instead of inventing new ones.

3.1 Anatomy

A notice carries nine elements, in this order.

1. Addressee. The recipient’s designated accommodation channel: human resources for an employer, disability services or the Section 504 coordinator for a school, the rights or compliance contact for a provider, the clerk or ADA coordinator for a court or public entity, and the accessibility or support channel for a platform. The notice goes to the function that applies the rule, never to the reader of the work.
2. Standing and means. A statement that the person uses AI as a disability tool, in the three prongs of the first companion paper and in the person’s own words: the barrier, what the AI does about it, and what that lets the person do that the person could not sustain without it (Puglisi, 2026b, Section 6.1). The validation of Section 4 is attached where one exists. No diagnosis is stated.
3. The policy or action at issue. The rule, the flag, the label requirement, the ban, or the decision, named and dated. The rule that restricts ongoing use is named. A score, flag, or label already applied is named and dated, and the request for its review runs separately from the accommodation request, which runs forward.
4. The protections it conflicts with. The provisions that reach this recipient, drawn from Section 2.5: for an employer, Title I’s accommodation and confidentiality duties; for a public entity, Title II’s modification and effective communication duties; for a European recipient, Article 9 of the GDPR and the national transposition of Directive 2000/78; in Spain, Article 7.6. The notice cites what applies and claims nothing that does not.
5. The request. That the person be held to the same standard by the different means; that no adverse action be taken on a detector score, a flag, or an impression until a named human has decided on evidence; that the accommodation process begin where one exists; and that, where the recipient’s process allows it, the policy not be applied to this use while that process runs.
6. The boundary. That the person’s use respects the recipient’s legitimate requirements on essential functions, data security, confidentiality, academic integrity, and the privacy of others, as the model clause names them. That where the work touches confidential or protected records the person asks for a configuration the recipient approves through its accommodation process, enterprise-governed, local, or without retention. That a worn device’s assistive function runs without retaining or publishing other people’s audio or video.
7. Confidential handling. That the notice be kept in a separate confidential file, that the recipient name who will see it, that it be used for the accommodation decision only, and that any other use the law does not require of the recipient needs the person’s written authorization.
8. Record. The date, the recipient, the method of delivery, and a request for written acknowledgment within the recipient’s ordinary accommodation timeline. Delivery is not receipt. The notice goes to the recipient’s accommodation channel or named officer, not a general inbox. The person keeps the date sent, the proof of delivery, who acknowledged it, and what came back. On a Union platform a complaint outside the Article 20 channel may not start the clock. In Spain a document addressed to a public administration is presented through its electronic register or the other channels the procedural law names, not by email (Ley 39/2015, Artículo 16.4). The person keeps the full record.
9. Reservation. That nothing in the notice waives any right, and that any further adverse action is taken with knowledge of the notice.

3.2 The model notice

The notice is a letter, and the paper offers its working text in the person’s voice. Each sentence can be used as written or adapted; the structure is what the paper proposes.

Before sending. This letter is a record, and records are seen again. The recipient will keep it, and a benefits agency may lawfully obtain it in a later review. The no-inference rule in Section 4.8 binds the issuer and the recipients who adopt the standard; it does not yet bind the Social Security Administration. A platform whose terms forbid AI content may treat the letter as an admission under those terms, and no located United States law stops it. State the third prong as what the tool lets you do that you could not sustain without it. Send the letter through the recipient’s accommodation channel and keep proof of delivery, and read Sections 4.8 and 7 before you sign. The notice starts no clock and stops none; the deadlines in Section 2.5 run from the adverse act.

Notice of Protected Assistive Use

To the designated accommodation channel of the recipient.

I use artificial intelligence as a disability tool. My disability-related barrier, what the AI does about it, and what that lets me do that I could not sustain without it, I state here in my own words, and I state no diagnosis. Where a validation under the AI Disability Standard exists, it is attached.

Your policy, system, or decision, which I name here, restricts or is about to restrict my ongoing use of that tool. Where an automated score, a flag, or a label requirement has already been applied to my work, I record its date at the foot of this notice and ask that a named person review it now on evidence.

That application conflicts with protections that already reach you. I assume you did not know this, because artificial intelligence as a disability tool is new to most institutions, and this notice is written to explain it as much as to record it. A reserved parking space exists because the law recognized a need and a placard tells the attendant who the space is for; a ramp exists because the law recognized that a locked step is a denial. This notice does the same work for my use of this tool. I name the protections here by statute, regulation, or article, as they reach you.

I ask that I be held to the same standard by this different means; that no adverse action be taken on a detector score, a flag, or an impression until a named person has decided on evidence; that your accommodation process begin where you have one; and that, where your process allows it, the policy not be applied to this use while that process runs.

My use of this tool respects your requirements on essential functions, data security, confidentiality, academic integrity, and the privacy of others. Where my work touches confidential or protected records, I ask to use a configuration you approve through your accommodation process, enterprise-governed, local, or without retention. Where my tool is worn, its assistive function runs without retaining or publishing the audio or video of other people.

Please keep this notice in a separate confidential file, tell me who will see it, use it for the accommodation decision only, and seek my written authorization for any other use that the law does not require of you.

This notice is dated and delivered as recorded at its foot. I ask for written acknowledgment within your ordinary accommodation timeline. I keep the full record.

Nothing in this notice waives any right. Any further adverse action is taken with knowledge of it.

The letter claims nothing the law does not already give. Where it asks for more than the law gives, as it does of a platform, it asks and does not demand. Its second paragraph is the plain-English request the Commission’s guidance recognizes. The three prongs carry the standing, and the validation, where one exists, stands where a medical file would otherwise be demanded. Its third carries two halves, the accommodation request that runs forward and the review request that looks back. A sender whose work has not yet been scored or labeled drops the second sentence. Its third and fourth together are the knowledge that triggers the duty: the conduct and the protection, named. The assertion that the application conflicts with the protections is the sender’s position, as the assertion in any cease and desist is. It stands or falls on the law the fourth paragraph names. The letter is written for a recipient who has never thought of AI as an accommodation, and its tone follows from that: it teaches before it warns. Its fifth is the same standard, different means, with the human checkpoint the record requires. Its sixth states the boundary the person’s own use keeps, which answers the legitimate requirements of the model clause before the recipient raises them. Its seventh is the ADA’s separate-file rule applied by the person to the recipient. Its last is what every cease and desist says. The provisions the fourth paragraph names are the ones in Section 2.5 that reach the recipient, and no others.

3.3 How the notice applies HIPAA’s principles

HIPAA’s minimum necessary standard requires covered entities “to limit protected health information to the minimum necessary to accomplish the intended purpose” (45 C.F.R. §§ 164.502(b), 164.514(d)). The notice applies that limit to itself. Its body states the three prongs in the person’s own words: the barrier, what the AI does about it, and what that lets the person do. It leaves out the diagnosis and medical detail unless the accommodation process asks for documentation, and then only what that process needs. It asks the recipient to name who will see it and to keep it in a separate confidential file, following the employer rule’s model. It states that it may be used only for the accommodation decision, and that any other use needs the person’s written authorization. The person keeps the full record.

3.4 The counter-instrument the notice must answer

Notre Dame’s engineering handbook requires the opposite of private notice: a sentence printed in the dissertation naming the accommodation and the office that approved it (University of Notre Dame Department of Electrical Engineering, 2025). That rule converts an accommodation into a permanent public label on the work, which is the third punishment in Section 1.4. The ADA confidentiality rule the paper borrows is an employment rule, and a dissertation is a student record. The law that reaches the department is Section 504, Title II or Title III, and FERPA, and no court has tested a compelled statement against them. The paper’s design keeps the notice where the ADA’s confidentiality rule keeps medical information, in a separate file seen by the people who need it. The dissertation says nothing.

3.5 Two registers: the first notice and the cease and desist

The letter in Section 3.2 is written in the form of a cease and desist because that form makes the recipient knowing and keeps the evidence. It is titled a Notice of Protected Assistive Use so that it reaches the accommodation channel (Section 2.1). A letter that opens by recording a conflict is likely to be routed to counsel wherever counsel exists. A request in the same elements is likelier to reach the officer who can grant it. That reading came from the review of the fourth draft, from readers taking the part of an accommodation officer, a litigator, and platform counsel. The paper carries both registers as the standard’s options and asks the community which works. It has no outcome data on either.

The first-notice variant keeps every element of Section 3.1 and changes the third and fourth paragraphs of the letter. In place of “Your policy, system, or decision, which I name here, restricts or is about to restrict my ongoing use of that tool,” the sender writes: “I request review of your policy, system, or decision, which I name here, under the protections that reach you, because it restricts or is about to restrict my ongoing use of that tool.” In place of “That application conflicts with protections that already reach you,” the sender writes: “I believe those protections reach this use, and I name them here so that you can review the policy against them.” Everything else stands: the standing in the person’s own words, the request for the same standard by the different means, the human checkpoint, the boundary, the confidential handling, the record, and the reservation. The reservation is what keeps the first notice a notice; the recipient is knowing from the date of delivery in either register.

When each fits. The first notice fits where the recipient has an accommodation channel and no adverse act has yet occurred. The rule is on the page, the person’s use is ongoing, and the question is whether the institution will treat the use as an accessibility question once asked. The cease and desist fits where an act has occurred, a score has been applied, a grade or a post has been penalized, or a prior request went unanswered within the recipient’s own timeline. A sender may begin in the first register and move to the second. The two letters together are the record Section 7 describes: notice, correction or its absence, and the harm that stands or grows. Question 28 asks which register produces the faster and fairer response, and whether the answer differs by recipient class.

4. The Validation

The notice says the person has standing. The validation is what lets the recipient believe it without opening a medical file, and it is the part of the design that the parking placard supplies. Disclosure inside disability systems is already accepted practice. A parking placard rests on a licensed physician’s certification (23 C.F.R. § 1235.4(b)), and a Social Security disability determination rests on medical evidence an agency holds (Social Security Administration, n.d.-a). An ABLE account rests on a certification signed under penalties of perjury with a physician’s diagnosis kept on file by the account holder (26 C.F.R. § 1.529A-2). A TSA Notification Card goes discreetly to one officer (Transportation Security Administration, n.d.). Each limits what is disclosed: the placard shows entitlement, the agency holds the determination, the certification stays on file, and the card describes a condition to one person. The paper borrows those forms. It uses kinds of disclosure that people with disabilities already live with, and that institutions already honor, instead of inventing a new one. This paper calls the instrument a validation. The sources it borrows from call theirs a credential, a card, a placard, or a certificate; where Section 8 quotes those sources, their word names the same thing.

Two more facts favor that starting point. First, the systems are graded and dated. The parking rule caps temporary placards at six months and requires renewal (23 C.F.R. §§ 1235.5(b), 1235.4). Second, they already reach across institutions. The federal guidelines ask each state to recognize placards “issued by other States and countries,” and every state does (23 C.F.R. § 1235.8). ABLE and vocational rehabilitation already treat an SSDI award as sufficient proof of disability (ABLE National Resource Center, 2026; 34 C.F.R. § 361.42). A validation that borrows from them inherits both properties as design goals. The graded expiry is the issuer’s to set, and recognition across institutions rests on adoption, which question 7 asks about, not on the placard rule. The validation is optional evidence of standing within the standard. No one needs it to request an accommodation, and the standard lets no recipient demand it as the price of considering one. A person who declines it keeps every protection the law gives.

4.1 Eligibility: the Social Security model

The Social Security Administration defines disability as “the inability to engage in any substantial gainful activity (SGA) by reason of any medically determinable physical or mental impairment(s) which can be expected to result in death or which has lasted or can be expected to last for a continuous period of not less than 12 months” (Social Security Administration, n.d.-a). The definition governs SSDI and adult SSI alike, and it covers physical and mental impairment on one footing. It defines no one’s rights under the ADA; Section 2.2 states the ADA’s own definition, and the two serve different jobs. The adult mental listings span 11 categories, including trauma- and stressor-related disorders, autism spectrum disorder, intellectual disorder, and neurodevelopmental disorders (Social Security Administration, n.d.-b).

The functional core of the mental listings is four areas the agency calls “the areas of mental functioning a person uses in a work setting”: understand, remember, or apply information; interact with others; concentrate, persist, or maintain pace; and adapt or manage oneself. The agency rates each on a five-point scale: none, mild, moderate, marked, and extreme. The paragraph B criteria of the mental listings are met by extreme limitation of one area or marked limitation of two. Five listings add a paragraph C route, and a person whose impairment meets no listing can still be found disabled at the later steps of the sequential evaluation (Social Security Administration, n.d.-b; 20 C.F.R. § 404.1520). The agency states that people use “the same four areas of mental functioning in daily activities at home and in the community” that they would use at work (Social Security Administration, n.d.-b). The standard’s domains therefore cover both settings, and no person’s path depends on a relationship to employment.

What the standard takes. The four areas, with physical function, become the standard’s functional domains. The companion paper’s three-prong functional test then decides whether a particular AI use answers the limitation (Puglisi, 2026b, Section 6.1).

Where the standard departs. The agency’s bar is unemployability: “We pay only for total disability” (Social Security Administration, n.d.-c). The standard’s bar is limitation in normal work, set below the listing level. The test for whether a use belongs under the standard is one question: does the use fit a purpose that offsets a disability, physical or mental. The clear case needs no argument. A person who cannot see uses AI to read, and the tool does what a reader would do. The gray area is the person who can do what most people do and cannot do it at the level counted as normal. A person who cannot sit or stand for more than an hour or two can work, and cannot hold a working day without a tool that carries part of it. The same holds for neurodivergence, post-traumatic stress, anxiety, and learning disabilities. For that gray area the measure of normal is the full working day, and the validity of the use slides with how far the person falls below it without the tool.

The gate and the comparator. The Social Security Administration supplies the gate and the methods the standard reuses. Its definition is the most severe test in the system. When the agency assesses residual functional capacity at the later steps of its evaluation, it ordinarily measures the capacity for sustained work as “a regular and continuing basis,” meaning “8 hours a day, for 5 days a week, or an equivalent work schedule” (Social Security Administration, 1996). The ruling carries its own exception for past part-time work at step four. The standard borrows that sustained-function convention and not the agency’s eligibility threshold. At that gate sits SSDI and the automatic qualifier of Section 4.2. Automatic qualification settles standing and nothing else; the use still has to meet the test. The scale does not stop at the gate. John Doe holds an eight-hour day, and holds it only by spending what others keep for the rest of their lives. The regulation names those costs as the measures of a limitation: the “difficulty, effort, or time required,” the “pain experienced,” and “the length of time a major life activity can be performed” (29 C.F.R. § 1630.2(j)(4)(ii)). The person beside him works sixteen and goes home. The eight hours are inside the definition; the sixteen are inside the statute’s purpose and outside its definition, and the paper says so. Jane Doe can work, and her ADHD keeps her from competing with people who do not have it. Both use AI as a disability tool, because the test asks whether the use offsets the disability in the pursuit the person has chosen. It does not ask whether the person can clear the minimum to be employed. The tool that lets John work the sixteen if he chooses is the same tool that lets him work the eight without paying for it twice. No one who can work eight hours is refused it.

The comparator of the scale is the ADA’s. Its regulation defines a disability as an impairment that “substantially limits the ability of an individual to perform a major life activity as compared to most people in the general population.” It adds that the impairment “need not prevent, or significantly or severely restrict” the activity (29 C.F.R. § 1630.2(j)(1)(ii)). The statute’s findings name the purpose: “equality of opportunity, full participation, independent living, and economic self-sufficiency,” against discrimination that denies people “the opportunity to compete on an equal basis” (42 U.S.C. § 12101(a)(7), (a)(8)). The two measure different things, work incapacity and functional limitation. The standard uses each for what it measures: the Administration’s gate and methods, the ADA’s comparator and purpose. The scale runs from the first to the second.

The grounds in the agency’s text. The grounds for setting the bar below the listing level are in the agency’s own text. It rates each area by whether the person functions “independently, appropriately, effectively, and on a sustained basis” (Social Security Administration, n.d.-b). A person can meet a standard and still be unable to sustain it, or sustain it only at a cost others do not pay. AI that lowers the cost of meeting the same standard is same standard, different means. The companion paper’s third prong already counts an intervention that maintains a capability (Puglisi, 2026b).

The qualifier behind the functional path. Behind the functional path sits a diagnosis, as a physician’s certification sits behind a parking placard. It exists. The issuer sees it at certification and keeps the certification, the functional domain, and the clinician’s name and date. The person keeps the diagnosis, as the ABLE account holder does under Section 4.3, and it appears on no face and in no notice. No issuer holds a medical file. The standard chooses that gate knowing its cost, which Section 8.1 states at full strength. The people a documentation gate excludes are the people least served by diagnosis today.

The range of disability. Disability runs from functional, to partial, to complete, and the standard covers the full range. The mapping below is a proposal for public review, not a finding:

• Complete: SSDI entitlement, whether reached through a listing or through the later steps of the sequential evaluation, taken through the automatic qualifier in Section 4.2.
• Partial: people who work with limits, including beneficiaries working under the agency’s work incentives in Section 4.8.
• Functional: limitation in normal work below the agency’s levels, reached through the functional path.

The research found graded models elsewhere that could inform the three ranges. Veterans Affairs ratings run in 10-point steps, with total disability based on individual unemployability where one disability is rated 60 percent or more or combined ratings reach 70 percent with one at 40 percent (38 C.F.R. §§ 4.1, 4.16(a)). The three tiers of vocational rehabilitation eligibility (34 C.F.R. § 361.5) and every European system examined (Section 2.3) are graded as well. The threshold question itself goes to the community as question 1.

4.2 Automatic qualifier: SSDI, following the ABLE precedent

Under the standard, SSDI receipt qualifies a person automatically, and it is the automatic qualifier the paper proposes. The determinations the research found beside it are named below and in question 2 as candidates the community may add. Automatic qualification settles standing and nothing else; the use still has to meet the test in Section 4.1. The precedent is ABLE. An ABLE account treats receipt of SSI or SSDI as the first route to eligibility for a disability that began before age 46, the onset age in effect since January 1, 2026 (ABLE National Resource Center, 2026; Social Security Administration, 2026e). The regulation’s text still reads 26 at the sections cited, and the statute controls. Vocational rehabilitation does the same: a Social Security disability beneficiary is presumed eligible and “Considered an individual with a significant disability” (34 C.F.R. § 361.42(a)(3)(i)). At the end of September 2026, 6,958,832 disabled workers were in current pay (Social Security Administration, 2026f). The automatic path therefore reaches a known population through a determination the federal government already made.

Holds through work. ABLE treats the “entitled to SSDI” criterion as met for a person in the extended period of eligibility in suspended status, or for a person receiving extended Medicare coverage after losing SSDI due to work (ABLE National Resource Center, 2026). The standard proposes the same rule, so the automatic path does not end when a person returns to work. The standard’s rule governs the validation only; the Administration’s own rules on benefits are untouched.

Reach. SSDI pays individuals insured through contributions on their own earnings, which shapes who the automatic path reaches. The agency allowed 34.5 percent of disabled-worker initial claims in the most recent year reported, and 37.1 percent across all disabled-beneficiary categories (Social Security Administration, 2024, Table 61). Those limits are why the functional path exists, and why question 2 asks which other determinations should qualify automatically.

Candidates. Each determination below shows one thing and leaves another open, and the community may add any of them. SSI uses the same adult definition and needs no work record, so it carries the same functional finding for people the SSDI path misses. ABLE and vocational rehabilitation already accept it beside SSDI (ABLE National Resource Center, 2026; 34 C.F.R. § 361.42(a)(3)(i)). An IEP or Section 504 plan records a school’s finding of need and says nothing about adult function. A VA rating records a service-connected condition on a graded scale with a comparator of its own (38 C.F.R. § 4.1). An accommodation already approved by an employer or school records that one institution found the need and the means, which is the finding the notice carries forward. Vocational rehabilitation eligibility records a state agency’s finding of an impairment that is a substantial impediment to employment (34 C.F.R. § 361.42(a)(1)). A state disability card or hidden-disability marker shows whatever its issuer verified, which in Maryland’s case is nothing (Section 2.4). Any award is a snapshot of the date it was made, and the validation’s expiry is what keeps a snapshot from standing in for the present.

4.3 Certification for everyone else: the ABLE pattern

A person who does not receive SSI or SSDI can open an ABLE account by certifying, under penalties of perjury, to three things. The person has a disability meeting the agency’s criteria for marked and severe functional limitations, the disability began before age 46, and the person “has obtained and will continue to retain a copy of the individual’s diagnosis” signed by a physician (26 C.F.R. § 1.529A-2(e)(1)). The diagnosis is retained, not submitted. It must be produced to the plan or the Internal Revenue Service only on request.

The same regulation limits what the certification proves. Paragraph (e)(5), headed “Restriction on use of certification,” states: “No inference may be drawn from a disability certification described in this paragraph (e) for purposes of establishing eligibility for benefits under title II, XVI, or XIX of the Social Security Act” (26 C.F.R. § 1.529A-2(e)(5)). The precedent the paper borrows therefore recognizes a certification for one purpose and refuses it as proof for another. The standard inherits that limit along with the mechanism.

What the standard takes. Self-certification, with documentation kept by the person and produced only on request. The paper describes ABLE’s model as certification by the account holder backed by a physician-signed diagnosis kept on file, never as pure self-certification, because that is what the regulation says. The standard borrows the mechanism and not ABLE’s threshold. ABLE’s oath certifies marked and severe limitations at the listing level, and the content of the standard’s oath is set by whatever threshold the community chooses under question 1. Who may sign the diagnosis is question 3. The no-inference clause is also the model for the standard’s own rule toward the Social Security Administration in Section 4.8: a validation issued to invoke an accommodation is not evidence of capacity for work. That rule is the standard’s own, inspired by ABLE’s purpose limit. ABLE’s clause protects the government from a lenient certification and lends this rule no authority of its own.

4.4 The face of the validation: the parking placard system

Part 1235 of Title 23 is a set of guidelines to the states, and it supplies the validation’s mechanics (23 C.F.R. Part 1235).

What the standard takes. A validation that shows entitlement only: an identification number, an expiration date, and the issuing authority’s seal, with no diagnosis, on the model of the placard’s face (23 C.F.R. § 1235.2(d), (e)). It carries none of the placard’s symbol, for the reason given below. Certification at entry by a licensed physician, followed by periodic renewal (§ 1235.4(b)). A temporary tier, valid for the period the certifier sets and capped at six months (§ 1235.5(b)). Reciprocity as the model, under which each state recognizes placards and plates “issued by other States and countries” (§ 1235.8); for the validation, recognition would rest on the agreements question 7 asks about.

What the standard leaves behind. The eligibility criteria, which are built on the ability to walk; public display, since a placard must be viewable from the front and rear of the vehicle; and the International Symbol of Access. The placard is parked. It does its work in a lot, at a distance, and the person walks away from it. The validation walks in with the person, into the interview, the classroom, and the venue. Anything printed on it is read at arm’s length by someone who knows the holder’s name. So the standard borrows the placard’s mechanics, an issuer, an entitlement, an expiry, reciprocity, and no diagnosis, and not its public face. No disability symbol appears on the validation, and recognition comes from the standard’s name and from the notice the card accompanies. The symbol is the recognized mark for a parking space because enforcement reads it from a distance (23 C.F.R. § 1235.6). That is the job the validation does not have. Parking is physical, and it supplies mechanics, not eligibility.

Three layers of disclosure. Disclosure runs in three layers, and the standard requires only the first. The diagnosis is shown to the issuer at certification, as a physician’s certification is shown to the agency that issues a placard. The issuer keeps the certification and not the diagnosis, and the person keeps the diagnosis (Section 4.3). The validation shows the entitlement and nothing behind it. Beyond that, disclosure is the person’s choice: to an institution through the notice, to a reader, or to no one. The standard requires disclosure to no institution, platform, or reader at any point on the scale, and the sequence matters. As long as declaring AI use brings punishment, any disclosure rule is a tax on honesty paid by the people who followed it. The Open University acts on that when it tells disabled students they need not report AI use as a reasonable adjustment (The Open University, 2025). The punishment has to stop before the disclosure question can be asked fairly, and the paper asks it in question 27 rather than answering it.

The misuse record, carried as dissent. The California State Auditor found in 2017 an average misuse rate of 15 percent across the enforcement operations it reviewed, which is an enforcement sample and not a rate for placard holders at large. It also found that about 35,000 placard holders matched the Death Master File and that about 26,000 holders were over 100 years old (California State Auditor, 2017). Any validation invites that record, and question 8 asks how to deter false claims without moving the burden back onto genuine users.

4.5 Presentation: the TSA Notification Card

The TSA Notification Card lets a traveler “describe your health condition, disability or medical device to the TSA officer in a discrete [sic] manner.” It states on its face: “I understand that presenting this card does not exempt me from screening” (Transportation Security Administration, n.d.). Notice is voluntary and discreet, given to the officer applying the rule, and alternate procedures provide an equivalent level of screening. That is same standard, different means in working federal form.

What the standard takes. Discreet presentation to the person or system applying the rule, with no exemption from the standard. What it leaves behind. The card carries no certification behind it.

4.6 The institution’s side: the companion paper’s model clause

Nothing in this AI policy is intended to restrict the use of AI or AI-enabled functionality approved as a reasonable accommodation or other disability-related assistive technology. Requests for such use will be evaluated through the organization’s applicable disability accommodation process, subject to legitimate requirements concerning essential functions, security, confidentiality, academic integrity, and fundamental alteration. (Puglisi, 2026c, Section 16.3)

The clause guarantees no outcome. It guarantees that a request is treated as an accessibility question from the start. The Open University already runs the practice the clause describes. So does the United Kingdom’s Department for Environment, Food and Rural Affairs, which made full Microsoft Copilot licences available as a reasonable adjustment after trials showed benefits for neurodivergent and disabled colleagues (The Open University, 2025; Ede, 2026). Polk County Public Schools in Florida bars students from wearing smart glasses during the school day. The rule provides that the restriction does not interfere with technology a student is authorized to use under an individualized education program or a Section 504 plan (Polk County Public Schools, 2026). That is the clause in a device rule.

4.7 The record

Every channel of legal exposure the companion paper maps asks for the same artifact, “a record that can be produced” (Puglisi, 2026c, Section 14). In this setting that record is the accommodation file showing individualized review. The record captures decisions and their basis, never a detector flag or a human impression alone. The record may hold the flag as an unverified input, with the detector, its version, and the date. It never holds the flag as a finding. The named human deciding on evidence is the standard’s own safeguard. The law requires it in two places the paper maps, GDPR Article 22 where a decision is solely automated with legal or similarly significant effects and DSA Article 20(6) for complaint decisions. No other provision this paper maps requires it, so the letter asks for it rather than claiming it. Under Checkpoint-Based Governance, a flag is not a decision until a named human completes it on evidence. The framework’s canonical text states that “Human in the loop means the person is present and participating. It does not require that the person holds authority or answers for the outcome” (Puglisi, n.d.). Elish (2019) names the alternative a moral crumple zone, in which responsibility “may be misattributed to a human actor who had limited control over the behavior of an automated or autonomous system.” In the detector case the zone runs two ways. The disabled writer absorbs the blame for the detector’s error, and the teacher or moderator who acts on the score stands as the responsible agent with limited control. A record of decisions, kept on both sides, is the design response. A flag recorded as a finding would follow a person the way an unexplained employment gap does.

4.8 Do not punish the attempt: working, creating, and trying while disabled

The paper holds two positions firmly. A person who tests work, works part time, or works with limits should not lose standing for trying to reach what is called normal. And a person who finds a way to be productive, through a podcast, a blog, or artwork that may bring no profit or employment, should not be punished for it. That activity is part of a healthy human life.

The tension. The agency’s definition rests on the inability to engage in substantial gainful activity. AI that helps a person work, create, or publish could be read as proof that the person can work. Cleveland v. Policy Management Systems Corp. (1999) is the Supreme Court saying that a benefits claim and an accommodation claim can both be true. The Administration decides capacity without accommodation and the ADA asks about capacity with it, and the person must be able to explain how. The no-inference rule below is the design’s half of that explanation; the third prong, worded as Section 3 words it, is the person’s.

The agency already lets beneficiaries test work. The trial work period lets an SSDI beneficiary test the ability to work for at least nine months, which need not be consecutive and must fall within 60 months. Full benefits continue regardless of earnings so long as the work is reported and the impairment persists. A month counts when gross earnings exceed $1,210 in 2026, or when self-employment runs more than 80 hours in the month (Social Security Administration, 2026b; 2026c). An extended period of eligibility then pays benefits for 36 months in any month where earnings stay below the substantial gainful activity level. That level is $1,690 per month for non-blind individuals and $2,830 for statutorily blind individuals in 2026, and $1,770 and $2,950 in 2027 (Social Security Administration, 2026a). An unsuccessful work attempt that ends within six months because of the disability does not count against the person. Expedited reinstatement restarts benefits within five years, and Medicare continues for at least 93 months after the trial work period (Social Security Administration, 2026a; 2026b). These are SSDI rules; SSI’s work rules differ and are not described here.

The agency already separates hobbies from work, and leaves a door open. The regulation states that the agency generally does not consider activities “like taking care of yourself, household tasks, hobbies, therapy, school attendance, club activities, or social programs” to be substantial gainful activity (20 C.F.R. § 404.1572(c)). The door is in the next clauses. Work is gainful if it is “the kind of work usually done for pay or profit, whether or not a profit is realized” (20 C.F.R. § 404.1572(b)). Work that is not substantial gainful activity “may show that you are able to do more work than you actually did” (20 C.F.R. § 404.1571). A podcast, a blog, or artwork can fall in that space, which is why question 17 exists.

Why the activity matters to health. A World Health Organization scoping review of over 3,000 studies found a role for the arts in preventing the onset of mental illness, in promoting health, and in helping people experiencing mental illness (Fancourt & Finn, 2019). The review addresses the arts; extending it to podcasts and blogs is an extension the paper names as such.

The agency already counts supports and deducts assistive technology. In deciding substantial gainful activity the agency considers “extra assistance or employer accommodation to complete job duties,” such as a job coach. It uses only earnings that represent the real value of the work performed (Social Security Administration, 2026d). Impairment-related work expenses are deducted from gross earnings. The agency’s own list of deductible items includes “Assistive technology that people with disabilities use for employment-related purposes,” including “software applications, computer support services” (Social Security Administration, 2026d). Software unrelated to the person’s disability and employment is not deductible, so general-purpose AI used as an accommodation sits between the two, which is question 15. Two limits sit beside that. An impairment-related work expense deducts the cost of the tool, not the earnings made with it (20 C.F.R. § 404.1576). A subsidy reduces countable earnings only where the employer pays more than the work is worth (20 C.F.R. § 404.1574(a)(2)). Earnings a person makes independently with the tool count in full.

The limit. The agency can consider medical evidence of recovery at any time, so benefits can stop for medical improvement even during the trial work period. A person in the Ticket to Work program working toward an employment goal with a provider is not subject to a continuing disability review (Social Security Administration, 2026c).

The same activity on platforms. The podcast, blog, or artwork the agency might weigh is also the content that platform AI rules reach through distribution penalties and perception enforcement, the companion paper’s Types IV and VI. The second position applies in both places. What the standard takes. AI used as an accommodation is a support of the same kind the agency already recognizes in job coaches and assistive software. The validation records it as a support, so work done with it is read as work done with an accommodation. The validation adopts a no-inference rule of its own, inspired by ABLE’s purpose limit in Section 4.3: nothing in it is evidence of capacity for work. ABLE’s clause protects the government from a lenient certification and lends this rule no authority of its own. The rule binds the issuer and the recipients who adopt the standard. Whether it can bind the agency is question 14. Until it does, a dated record that AI is a person’s accommodation for work is a document the person should expect to see again in a continuing disability review.

5. The Stress Test: AI Glasses

5.1 Why this case

One device carries every strand of the paper at once: a general-purpose product used as assistive technology, a camera that bystanders fear, institutions that ban it, strangers who confront the people wearing it, a vendor that funds disability deployments and seeks assistive device status, and regulators on two continents reasoning about blind wearers as a class. The same sensor serves both sides. The camera that reads a menu aloud for a legally blind person is the camera that raises the fear of covert recording.

The glasses are also where the notice meets its hardest recipient. A courthouse door, a venue, and a cruise line apply a device rule to everyone. The person presenting a notice at that door is asking a rule-applier with no accommodation training to recognize a protection the rule did not mention. What the notice can and cannot do there is the test of the whole design.

5.2 The accessibility use

Meta AI on the Ray-Ban Meta glasses can describe aloud what is in front of the wearer, and a Detailed Responses setting gives fuller descriptions. A Be My Eyes call connects the wearer to a sighted volunteer (UploadVR, 2025). The vendor’s own terms say not to rely on the AI. That is the companion paper’s verification burden falling on a person who may be unable to verify a visual claim independently (Puglisi, 2026b). David Zucker of New York City told a reporter he was born with oculocutaneous albinism type 2 and is legally blind. He bought the glasses in late 2024 and used them to read menus and move through the city. In September 2026 he said: “I’ve really just stopped wearing them anywhere lately because of the radically altering public perception of them,” and “What I loved about them was they allowed me to exist anonymously” (Pavlou, 2026). Meta is donating the glasses to more than 130,000 legally blind United States veterans through the Blinded Veterans Association and partners. The American Council of the Blind runs a Meta-funded program with live demonstrations, open office hours, and individual support for blind users (Meta, 2026; Heaney, 2026; American Council of the Blind, n.d.-b). Meta’s vice president of wearables said the company is “very interested” in assistive-device classification by United States health agencies, with no agency, filing, or timeline named (Capelouto, 2026). The FDA cleared Meta’s hearing assist software, K254044, on July 13, 2026, and has cleared no vision function (U.S. Food and Drug Administration, 2026).

The disability community is not of one mind about the device. The National Federation of the Blind’s convention address warned that the AI “just makes things up” (Mosen, 2025), and an AFB review called the glasses “an accessibility tool only by coincidence” (Kelley, 2025). The Royal National Institute of Blind People in the United Kingdom said in September 2026 that “AI Smart Glasses can be a powerful assistive tool, but it’s vital that privacy and consent are built into their manufacture and use,” and that “Blanket bans risk excluding blind and partially sighted people who use AI smart glasses as an accessibility tool. Venues and service providers need to take an inclusive, proportionate approach that protects privacy while enabling assistive technology like AI smart glasses as a reasonable adjustment” (RNIB, 2026). Keri Gray, a disability advocate who has seen more blind people using the glasses, acknowledges the privacy risks and says “there does have to be regulation” (Capelouto, 2026). Vision Ireland argued that the glasses are assistive technology and reported inconsistent treatment of VAT refunds (Vision Ireland, 2025).

5.3 The privacy outcry

The record of misuse and complaint is long. Reports of non-consensual recording led United Kingdom venues including JD Wetherspoon, Soho House, ATG Theatres, and a restaurant group to bar filming with the glasses. Wetherspoon’s chairman said the glasses “seem to breach this code, and common sense, by enabling surreptitious surveillance” (VR.org, 2026). France’s 3018 helpline received about 20 reports of women filmed without their knowledge by September 8, 2026 and about 30 by October 6, after which the Paris prosecutor opened investigations. The association’s advocacy lead told AFP that “voir un voyant s’allumer ne veut pas dire que l’on consent à être filmée” (e-Enfance, 2026a; 2026b). In the Netherlands, the foundations Offlimits and The Firewall served Meta with a formal notice in September. They will ask the Amsterdam court on November 26, 2026 to halt sales of the glasses so long as covert recording remains possible (DutchNews, 2026; Offlimits, 2026). A woman covertly filmed by a stranger in the United States told CBS News “I had no say” and that the clip reached “probably over 200,000 views” (CBS News, 2026). Meta disabled photo and video capture on thousands of glasses after finding the recording indicator covered or destroyed. It estimated the affected units at under 0.1 percent of the 7 million sold in 2025 (Capelouto, 2026). In October 2026, owners reported that the glasses had recorded audio without the wake phrase. Meta said “This issue has been resolved” and gave no account of the cause (TechRadar, 2026; TechSpot, 2026).

The measured behavior matches the outcry. Security.org’s August 2026 survey of 1,096 adults included 103 owners recruited so the group was “large enough to study” and “not meant to represent all U.S. owners.” It found that “only four percent of owners say they always tell nearby people when they are recording,” that “nearly a quarter never disclose,” and that 85 percent of adults favor a law requiring a visible recording indicator. In the same survey, 25 percent of owners use the glasses for vision correction, and there is no disability breakdown of the 20 percent of owners who reported a negative public reaction (Security.org, 2026). In France, the CNIL found that 67 percent of adults see a privacy risk, while noting the devices’ usefulness “particularly for people with disabilities (visual or hearing assistance)” (CNIL, 2026).

5.4 Denial of access

Of roughly 20 verified device policies, one adopted rule and one consultancy draft state a disability path. The New York State Unified Court System’s amended memorandum of July 1, 2026, effective July 20, prohibits smart glasses in every court facility for “all individuals entering court facilities,” with no exception, exemption, or accommodation for disability, assistive technology, or medical need. Its sign advises prescription wearers to bring a regular pair of glasses. The court system’s general Americans with Disabilities Act accommodation process, which exists on its accessibility pages, is not referenced in the memorandum (New York State Unified Court System, 2026). No report of the ban located by this paper records a disability exception. The courts in England and Wales confiscate the glasses on entry, and MSC Cruises lists smart glasses among prohibited items in public areas (Legal Cheek, 2026; Younker, 2025). Hans Anders, with over 400 stores in the Netherlands and Belgium, stopped selling the glasses on October 2, 2026. Potsdam banned them in pools and saunas, and Norway proposed a temporary ban with no disability exception (TNW, 2026a; 2026b; CORRECTIV, 2026). Zucker stopped wearing the glasses at his gym (Pavlou, 2026). The two paths are a school district’s rule and a consultancy’s draft. Polk County Public Schools bars smart glasses during the school day while preserving technology a student is authorized to use under an individualized education program or a Section 504 plan (Polk County Public Schools, 2026). A Canadian school-safety consultancy’s draft policy carries an accommodation clause (The White Hatter, 2026). Both are the model clause in a device rule, and both require the person to raise it first, which is question 20.

A legal fact, reported as such. The one federal case located on a prescribed smart-glasses accommodation at work, Cedeno v. Walt Disney Parks and Resorts U.S., Inc., No. 6:25-cv-02046 (M.D. Fla.), ended without a ruling. The companion paper described it through its amended complaint. The docket shows a notice of voluntary dismissal on October 16, 2026 and an order dismissing the case on October 20, 2026. Whether a settlement preceded the dismissal is not stated in any public document (Cedeno v. Walt Disney Parks, 2026). The case yields no decision on whether prescribed smart glasses are a reasonable accommodation, and this paper draws none from it.

5.5 Hostility toward wearers

On a New York City subway in October 2025, a woman accused a man of recording her, took his Meta glasses, and smashed them. The wearer stated in his own video that he filed a report with the police and that the charge is a misdemeanor. Commentary online hailed the woman as a privacy hero, and the reporting reached only the wearer (Weedston, 2025; Tangermann, 2025). The reporting does not say the man used the glasses for a disability. What the case shows is the threat a wearer faces and the public opinion behind it. After privacy coverage grew, Zucker felt the stares and largely stopped wearing the glasses (Pavlou, 2026). A notice presented to a venue does nothing about a stranger on a train. The public is outside the system the paper proposes, which is question 22.

5.6 Regulators on both sides

The European Data Protection Board’s Statement 1/2026 of October 13, 2026 names “disability aid (for example by providing audio assistance and real-time hazard awareness)” among the devices’ uses. It places purely personal use under the household exemption while construing it narrowly, and it says the GDPR still applies to the producer and the AI provider. A footnote records that LED indicators “might be difficult or impossible to notice for individuals who are visually impaired” (European Data Protection Board, 2026). The Hamburg data protection authority’s technical review of the Ray-Ban Meta glasses reasoned that a visually impaired wearer’s legitimate interest can be assessed case by case. It added that such wearers “are typically easily recognizable from a distance as visually impaired by yellow armbands, a white cane, or a guide dog wearing a white harness, which can increase the transparency and predictability of other assistive devices” (HmbBfDI, 2026). That reasoning prizes public visibility of the aid, and it sits against a private validation. Section 9 puts the tension to the community as question 18. The German data protection conference endorsed the Hamburg report on September 23, 2026 and stated that for consent “Eine „Hinweis-LED” genügt hierfür nicht.” It acknowledged that smart glasses “beispielsweise blinde Menschen in ihrem Alltag unterstützen,” and it called on the German or European legislature to strike a balance in public space (Datenschutzkonferenz, 2026). The Italian Garante republished the EDPB statement on October 14, listing assistance to persons with disabilities among the uses (Garante per la protezione dei dati personali, 2026). The United Kingdom’s Information Commissioner’s chief executive wrote, “As a severely visually impaired person,” that “Most transparency mechanisms in physical spaces rely heavily on visual cues … The transparency mechanisms society came to rely on largely assumed that people could see them” (Information Commissioner’s Office, 2026). No United States agency has classified camera-based AI glasses as an assistive device. No CMS, FTC, or DOJ statement on the question was found as of October 27, 2026.

The one United States legislative text that paired a recording-indicator mandate with a disability exception, California’s SB 1130, was vetoed on September 30, 2026. The Governor’s message gives two reasons: that the bill “defines several terms too broadly or imprecisely” and that “existing law already makes it a crime to use an electronic device to record a person in any area in which the occupant has a reasonable expectation of privacy.” It does not mention the bill’s exemption for assistive technology used under the ADA (Office of the Governor of California, 2026). Pennsylvania’s HB 2603 would require a visible recording indicator with no disability exception (Pennsylvania House of Representatives, 2026).

5.7 What the case tests

The glasses test six things. They test the companion papers’ thesis of a consumer device working as assistive technology; the companion paper’s Type I use prohibition carried from software into physical space; whether a device ban carries an accommodation path; the difference between a camera seen, which is a reading of intent, and a recording, which is a fact; vendor dependency in a deployment the vendor funds and can switch off remotely; and the legitimate counter-interest of bystander privacy, which the model clause’s security and confidentiality limits already name. One more fact narrows the case. Waveguide display glasses that caption speech on the lens with no camera now exist for Deaf and hard-of-hearing users, as a vendor’s own page describes (AirCaps, 2026). The captioning case and the camera case can therefore be separated, and the validation’s hardest work is on the camera side.

What the notice does at the door. Presented to a court, the notice is a written request for a modification that Title II requires the entity to consider, subject to its legitimate safety requirements. For the reading and captioning functions it is also an effective communication request to which the entity must give primary consideration (28 C.F.R. §§ 35.130(b)(7)(i), 35.160). The New York memorandum’s silence on accommodation does not remove that duty, and a dated notice records that the court knew. The defense the law gives a court or a venue is direct threat, a significant risk to the health or safety of others judged by an individualized assessment on the best available objective evidence (28 C.F.R. §§ 35.139, 36.208). Bystander privacy is a real concern, and it is not a health or safety threat in that sense. It is argued under the general modification rule, where the entity must still consider the request. Presented to a private venue, it has force where Title III or the national law reaches the venue and builds a record where it does not. For an auxiliary aid, Title III asks the venue to consult the person and leaves the ultimate choice to the venue, provided the method chosen is effective (28 C.F.R. § 36.303(c)(1)(ii)). Title II’s primary consideration does not carry over. Presented to no one, on a train or in a gym, it does nothing, which is why question 22 exists. And against a device that records without command, as the Ray-Ban Meta glasses did in October 2026, neither the notice nor the validation answers the bystander. That is question 24.

A worked decision. A venue that receives the notice separates four things. Live description with nothing stored is the assistive function, and the EDPB’s storage-off default for real-time assistance describes it (European Data Protection Board, 2026). Recording or streaming is the conduct bystanders fear, and the venue’s rule on it stands. A less intrusive means may exist, such as captioning-only waveguide glasses for a Deaf wearer (AirCaps, 2026) or a human reader, and under Title III the venue may choose it where it is effective. Where the door cannot decide, the question goes to the venue’s accommodation officer, which is the channel the notice names. The notice promises no outcome at any of the four points; it makes the venue decide them in that order and on the record.

6. The Instrument Assembled: Notice plus Validation

The components fit together in eight steps. Each step names the system it comes from, so that the reader can see what is borrowed and what is new.

1. A person meets the eligibility standard in one of three ways: automatically through an SSDI award, as ABLE already accepts it; through another determination the community adds under question 2; or through the functional path at a threshold below unemployability that the community sets (Sections 4.1 and 4.2).
2. A person on the functional path certifies under penalty of perjury, as the standard proposes and as the ABLE account holder already does, and keeps the diagnosis. A no-inference rule of the standard’s own, modeled on ABLE’s, is carried into the certification (Section 4.3).
3. An issuer provides a validation, to a person who asks for one, that shows entitlement only: an identification number, an expiration date, and the issuer’s mark, and no diagnosis, as the parking placard and the German, French, Dutch, and Spanish credentials do (Section 4.4). No one needs the validation to send the notice.
4. The person sends the notice of Section 3, in the register Section 3.5 fits to the case, to the recipient’s accommodation channel or named officer, and keeps proof of delivery. The validation is attached where one exists, under the HIPAA principles of Section 3.3 and, in the Union, under Article 9 of the GDPR, as the TSA card is presented at the checkpoint and the French RQTH is disclosed at the worker’s option (Section 4.5).
5. The recipient, now knowing, holds the person to the same standard and decides on the different means through its accommodation process, granting it or an equally effective alternative where the law requires, which is the companion paper’s model clause (Section 4.6). Where no law compels the recipient, the notice stands as the record of its knowledge (Section 2.5).
6. Both sides keep a dated record of the decision and its basis, with the record of accept, modify, or reject that Checkpoint-Based Governance supplies, and no flag or reading is recorded as a finding (Section 4.7).
7. Other institutions recognize the validation as practice, on the model of the parking guidelines, so that the next notice carries the same attachment and the person does not start over (Section 4.4; question 7).
8. A person who returns to work, or who creates and publishes without pay, keeps the validation through the Social Security Administration’s work incentive periods and under the do-not-punish rule. The validation is not evidence of capacity for work within the standard, and Section 4.8 states that rule’s limit toward the Administration.

Steps 4 through 6 are available to any person today, before any issuer exists. The three prongs in the person’s own words carry the standing, as the model notice is written. What the person lacks today is step 3, a validation any recipient will take at face value, and step 7, recognition of it by the next recipient. Those two steps are what the paper asks the community to design, and neither creates a right the person does not already hold.

7. What the Notice and the Validation Do Not Do

The notice and the validation create no right. Every protection the notice names exists before the letter is written, and the letter’s work is to make the recipient knowing. Before it, the institution could say it did not know that the tool was a disability tool or that its rule reached a protected use; after it, it cannot. That is what a cease and desist does, and it is all this instrument claims to do.

The sequence is notice first, with the explanation the recipient may need; correction, where the recipient acts on what it now knows; and the record, where the harm stands or grows. The record then shows three things in order: the right, how the person qualified for it, and the notice that was given, so that no one can later say they did not know.

The instrument exempts no one from the standard. The TSA card states the same limit on its face, and the model clause states it in its first words (Puglisi, 2026c).

It does not cover substitution. The first companion paper’s excluded case, a report written by AI on a subject the person never studied and submitted unread, fails the functional test. No notice reaches it (Puglisi, 2026b).

It does not treat. The first paper separates offering therapy from organizing communication during psychological stress, and the notice concerns the second only.

It labels no one in public. The validation is presented privately, and the notice travels to the enforcing function, not to the reader. Where an institution requires a public statement of the accommodation, as the Notre Dame Electrical Engineering dissertation rule does, the instrument holds that the private notice is the standard. The public statement is the exception that the institution must justify (Section 3).

It does not record a flag as a finding. A detector score and a human impression are readings, and the record holds decisions by named humans on evidence (Section 4.7).

It does not make refusal unlawful. The law lets a recipient refuse where the means would impose undue hardship, fundamentally alter the program, pose a direct threat to health or safety, or breach a duty to others. A hospital that bars a recording tool from patient rooms, an exam that measures unaided recall, and a firm that keeps client files off a public model are examples. Under the ADA the recipient carries the burden of showing undue hardship, fundamental alteration, or direct threat on evidence; a duty to others is argued under the general rule, and the burden varies by statute and forum. The notice’s work is to make that showing happen rather than to make it unnecessary.

It does not by itself protect the sender from retaliation. The ADA bars retaliation against a person who requests an accommodation or opposes a practice the Act forbids (42 U.S.C. § 12203). A notice is such a request, and an adverse action that follows it is the first fact a retaliation claim records. Outside the statutes that carry that bar, the record is the protection.

It makes no claim about medical recovery. That determination stays with the Social Security Administration under its own rules.

It never treats creative or expressive work as a finding against the person within the instrument. A podcast, a blog, or artwork is part of a healthy life, not evidence of recovery, and the instrument takes the position in Section 4.8 as its own. The Administration’s own evaluation is untouched and is the subject of question 14.

It does not reach the public. A notice works with institutions and platforms. It does nothing on a train, in a gym, or on a street, and the paper says so in Section 5.5 and asks the question in Section 9.

It binds no platform and no publisher. Where no statute compels the recipient, the notice builds a record, and in the Union it opens the statement-of-reasons and complaint route the Digital Services Act already gives. The paper claims nothing more, and says so in Section 2.5 rather than in a footnote.

It grants no one exclusive access to the tool. The equalizing the paper describes in Section 1 comes from removing the penalty, not from reserving the tool. Where the tool is open to all, the disabled user’s position is restored, not raised.

It makes the validation no one’s prerequisite. The validation is optional evidence of standing within the standard. No one needs it to request an accommodation, the standard lets no recipient demand it as the price of considering one, and a person who declines it keeps every protection the law gives (Section 4).

It creates no legal duty. Every duty it invokes exists already. Every limit on those duties, from the circuit split on online platforms to the HIPAA coverage gap to the exclusion of employment from the European Disability Card, is carried in Section 2. The author is not a lawyer, and this paper is exploratory. It claims no legal effect for the instrument itself, and where the notice has an effect, the effect is the law’s, attached to a request the law already recognizes.

8. The Case Against the Instrument

A paper that proposes a notice for disabled people, with a validation layer behind it, owes its readers the full case against both. The sources below speak of a credential, a card, or a registry. In this paper’s design those words name the validation of Section 4, and the objections are carried against it at the strength their authors gave them. Every objection below is left standing. The paper answers none of them here; it puts them to the community in Section 9.

8.1 Against the eligibility model

The Supreme Court held in Cleveland v. Policy Management Systems that the Social Security Administration “does not take the possibility of ‘reasonable accommodation’ into account” (Cleveland v. Policy Management Systems Corp., 1999). A person who claimed total disability for benefits must therefore explain the claim before asserting that she could work with accommodation under the ADA. The Administration says of its own program, “We pay only for total disability,” and “No benefits are payable for partial disability” (Social Security Administration, n.d.-c). It allowed 34.5 percent of disabled-worker initial claims in the most recent year reported, and 37.1 percent across all disabled-beneficiary categories, and processing averaged between 184 and 231 days with state ranges from 108 to 452 days (Social Security Administration, 2024, Table 61; Bisignano, 2026; Disability Exchange, 2026). The Congressional Research Service states that the programs “do not pay benefits for partial or short-term disabilities” (Congressional Research Service, 2025). The Center on Budget and Policy Priorities describes United States eligibility as among “the most stringent” among advanced economies (Center on Budget and Policy Priorities, 2025). A credential built on that definition inherits a system that rejects most applicants, that treats disability as binary, and that was never designed to measure a functional barrier at a threshold below unemployability. The Government Accountability Office found unexplained racial gaps in allowance rates, especially at the hearing level, and a 2025 Census Bureau study found the gaps “generally not statistically significant” after adjustment (U.S. Government Accountability Office, 2002; Mitchell & Thompson, 2025). The dispute is live, and the paper records it as one.

Scholars attack the normal-work axis itself. Jamshed and colleagues warn that generative AI “can reinforce the normative value of productivity,” and Hamraie and Fritsch write, “we agitate against independence and productivity as requirements for existence” (Jamshed et al., 2025; Hamraie & Fritsch, 2019). Sins Invalid holds that “Our worth is not dependent on what and how much we can produce” and that “state solutions inevitably extend into further control over lives,” and Degener holds that “human rights do not require a certain health or body status” (Sins Invalid, n.d.-a; n.d.-b; Degener, 2016). On that view, a credential keyed to capacity for work measures the wrong thing and concedes the wrong premise.

8.2 Against certification and documentation

Macfarlane’s objection is that under current doctrine “employees are not disabled … until a doctor says so,” and that documentation requirements “distinguish between those who can obtain medical documentation and those who cannot” (Macfarlane, 2021; 2025). The population that perception enforcement harms most is the population a documentation gate excludes. McCrossin’s single-practice model estimated that 80 percent of autistic females were undiagnosed at 18, and a 2026 King’s College London study put ADHD diagnosis in England at 1.19 percent against an expected 3 to 5 percent. The CDC reported in 2024, and again in December 2025, that 15.5 million United States adults had an ADHD diagnosis in 2023, 55.9 percent of them first diagnosed as adults (McCrossin, 2022; King’s College London, 2026; Staley et al., 2024, as carried in Ashman et al., 2025). Disability Belongs describes being “forced to translate our lived experiences into paperwork” and verification that “often becomes suspicion” (Disability Belongs, 2026). Disability Rights UK wrote of a system that “assumes we are fraudulent until proven innocent,” and Amnesty UK of one that “reduced people to data points” (Disability Rights UK, 2021; Amnesty International UK, 2025). Anderson’s argument, as the Stanford Encyclopedia carries it, is that requiring citizens to establish an inferior endowment is disrespectful. Wolff’s is that status enhancement “does not require us to identify particular individuals,” which saves “the costs of identifying and excluding some individuals” (Putnam et al., 2026). The European Disability Forum stated in 2022 that “it should never be an obligation to hold such a Card to prove ones’ disability to avoid discrimination and stigmatization,” and that national systems take “a purely medical approach” (European Disability Forum, 2022). The CRPD Committee’s General Comment No. 6 holds that the accommodation duty is “not limited to situations in which the person with a disability has asked for an accommodation,” and that the burden of proof rests on the duty bearer (Committee on the Rights of Persons with Disabilities, 2018). Spain’s new law protects the person “con independencia de la existencia de reconocimiento oficial” (BOE, n.d., Article 64.2). On each of these views, a verification-first design is the wrong design, and the person should be protected without a card.

8.3 Against the credential and the record

The parking placard the paper borrows is the credential most studied for abuse. The California State Auditor found an average misuse rate of 15 percent across the enforcement operations it reviewed, an enforcement sample rather than a rate for holders at large. It also found 35,000 placard holders matching the Death Master File and 26,000 holders over 100 years old (California State Auditor, 2017). A policy analyst wrote that the abuse “undermines public support and sympathy” (Greenhut, 2017). The registry a validation could imply is the objection raised with the most history behind it. The Autistic Self Advocacy Network, the ACLU, and 80 co-signers wrote in May 2025 of “a long and troubled history with governmental efforts to find and track disability,” and that inadequate protections “may chill individuals’ willingness to … openly identify as autistic” (American Civil Liberties Union et al., 2025). ASAN’s Taylor-Parker said that past tracking “has ended in murder,” and the Lemkin Institute warned of “the danger of any state creating registries and lists of people whom it finds to be burdensome” (Hawkins, 2025; Lemkin Institute for Genocide Prevention, 2025). The Department of Labor incident of 2026 is the objection in working form. Functional-limitation categories, supervisor names, and job titles left a federal accommodation office, and a union official said the agency “can’t protect its employees” (Heckman, 2026b). The Electronic Frontier Foundation holds that “AI should never be part of a decision about whether a person will be … denied access to important government benefits like disability payments or medical care” (Electronic Frontier Foundation, 2026). The National Federation of the Blind resolved in 2025 against consolidating blindness services into a general disability agency, which cuts against a cross-disability gate and a generic symbol (National Federation of the Blind, 2025). The European Disability Forum reported that the Union’s own card lacks a tactile cue and a text alternative for its encoded data. A card built for disabled people can therefore be inaccessible to them (European Disability Forum, 2026a).

8.4 Against private presentation and notice

The notice rests on confidentiality, and the objection is that confidentiality fails where it is needed most. HHS states that HIPAA “does not protect your employment records” (U.S. Department of Health and Human Services, n.d.). In the Union, presentation of a disability credential is special-category processing by the recipient, which needs a lawful basis the recipient must be able to show (Regulation (EU) 2016/679, Article 9; Data Protection Commission, n.d.). The Center for Democracy and Technology and the American Association of People with Disabilities write that users of assistive tools “must often disclose sensitive, disability-related information in order to use them,” a forced trade-off (Aboulafia & Claypool, 2025). The Center’s earlier report found it “very hard, if not impossible” to avoid revealing status, with “a chilling effect” (Luria et al., 2024). Lydia X. Z. Brown writes that disabled people “are more likely to experience surveillance” and are “simultaneously hyper-surveilled and often erased” (Brown, 2021; Center for Democracy and Technology, 2022). Pierrès, Darvishy, and Christen found that only three of 33 disabled students could imagine ChatGPT as an accommodation on its own. They warned that “Reducing GAI to an accommodation could force students to disclose their disability status to access it” (Pierrès et al., 2025). Taheri and colleagues found that disclosure to the tool drifts “into contexts where it did not belong,” which the notice, addressed to the institution, does not govern (Taheri et al., 2026). In Germany, co-determination under § 87(1)(6) of the Works Constitution Act turns on “die objektive Eignung der Technologie zur Überwachung, unabhängig von der Absicht des Arbeitgebers,” so that a private credential cannot override collective law (Bitkom, 2025). The American Council of the Blind’s own text-to-speech guidelines hold that “Audiences must be informed when TTS is used” (American Council of the Blind, n.d.-a). That is a disability organization’s disclosure principle, and it cuts against treating every AI-origin disclosure as harmful. And the Notre Dame Electrical Engineering rule requires the public statement the notice is designed to avoid (University of Notre Dame Department of Electrical Engineering, 2025).

8.5 Against the premise that AI is assistive technology

The disability community’s own doubts about the tool are the strongest attack on the premise. The National Federation of the Blind’s convention address said the AI “just makes things up” (Mosen, 2025). The American Foundation for the Blind called the glasses “an accessibility tool only by coincidence” and asked “how can they trust that the description is factual?”, and its research found “mistakes that are not obvious to its users” (Kelley, 2025; American Foundation for the Blind, 2026; Silverman et al., 2026). The Autistic Self Advocacy Network tested plain-language tools and found that “all of them made big mistakes that changed the meaning of the text,” concluding that plain language “should be left to people, especially disabled people” (Autistic Self Advocacy Network, 2025). The National Council on Disability finds that “AI systems often replicate and amplify existing biases” and that there are “no comprehensive federal standards” requiring accessible AI (National Council on Disability, 2025; 2026). Wang, Kameswaran, and Kacorri catalogued 48 patterns and 175 instances of harm, including “Inability to verify algorithmic output” (Wang et al., 2025). Cha and colleagues found that “the most difficult questions receive the most confident answers,” and Johnson, Lewis, Mankoff, and Banner describe accessibility taxes and a “verification catch-22” (Cha et al., 2026; Johnson et al., 2026). Lamb, Brown, and Grossman found disabled writers “markedly more pessimistic” about generative AI, and Harstone describes AI as “cover for hostile remarks” (Lamb et al., 2024; Harstone, 2024). Boutin-Cooper and Puglisi frame AI refusal as a disability-justice choice (Boutin-Cooper & Puglisi, 2026). Frankowska-Takhari describes “moral compression” of disabled testimony by AI summarizers (Frankowska-Takhari, 2026). Seung and Basham find that generative AI “has the potential to bypass essential cognitive processes and undermine learning when students are overreliant” (Seung & Basham, 2026). Uebel and Hamamra hold that “‘AI authorship’ is a category mistake” (Uebel & Hamamra, 2026). Barandiaran and Pérez-Verdugo argue that generative AI has its own “generative locus,” which makes the extended-mind account of authorship weaker for a language model than for a screen reader (Barandiaran & Pérez-Verdugo, 2025). Gilly describes “accommodation repricing,” the price held constant while “The quantity behind it was metered,” which is the vendor dependence the companion paper names (Gilly, 2026). The European Disability Forum’s 2023 board resolution was “Concerned that persons with disabilities are at increased risk of becoming victims of fraud or misinformation through misuse of generative AI” (European Disability Forum, 2023). The National Federation of the Blind’s 2026 resolutions ask that AI tools be usable nonvisually and produce accessible code by default, and that AI interview platforms stop scoring “eye contact, facial expression, or other visual behavior metrics” (National Federation of the Blind, 2026). The organization’s live concern, that is, is AI used on blind people and the accessibility of the tool itself, neither of which a credential secures.

8.6 Against a credential overriding institutional rules

The ICMJE and COPE hold that AI “cannot be responsible for the accuracy, integrity, and originality of the work” and “cannot take responsibility” (International Committee of Medical Journal Editors, n.d.; Committee on Publication Ethics, 2023). A credential changes nothing about who answers for the work. Employment counsel ground AI restrictions in confidentiality, trade secrets, and privilege: “What you give generative AI, you can’t get back” (Society for Human Resource Management, 2023; 2026). Taylor & Francis requires every generative AI use disclosed with no exemption. The Journal of Oral and Maxillofacial Surgery’s author guide, inside a publisher whose policy carries an exemption, states none (Taylor & Francis, 2025; Journal of Oral and Maxillofacial Surgery, 2026). The College Board’s AP policy carries no accommodation exception on its page and gives a zero on uncompleted checkpoints (College Board, 2026). The University of Arkansas at Fort Smith limits an AI accommodation to “the specific accommodation intended and not available for broader use,” and Bologna requires the certification deposited first and states that reductions of program or exemption from parts of an exam are never possible (University of Arkansas at Fort Smith, n.d.; Università di Bologna, n.d.). A Berlin English department excludes AI even for “sprachlich-stilistische Überarbeitung,” and a Belgian university college treats forbidden generative AI use as an irregularity when communicated in advance (Freie Universität Berlin, n.d.; Arteveldehogeschool, 2026). The competence-standard objection runs through all of these. A rule can be about what the person must be able to do, and a different means is then not the same standard. A student survey found 55 percent saying AI on job applications gives “an unfair advantage” (BestColleges, 2023). Cruise lines and venues write device rules that protect third parties and address prescription wearers only by a backup pair (Royal Caribbean, n.d.; Younker, 2025; UK Cinema Association, 2023). Keri Gray, who advocates for the glasses, says “there should be places and moments that can be sacred to people” (Capelouto, 2026). The Center for Democracy and Technology’s Aboulafia warns of accessibility used as a “Trojan horse” for facial recognition (Aboulafia, 2026), and an Australian wearers’ group rejects a disability exemption as the solution: “Punish misuse,” “Do not ban independence” (Smart Glasses Australia, 2026). The Dutch foundations’ suit would halt sales of the glasses altogether, which leaves no exception to hold (Offlimits, 2026). The veto of SB 1130 removed the one legislative text that paired an indicator mandate with a disability exception. The Governor’s message does not address the exception, so the paper records the loss of the text and attributes no view on the exception to him (Office of the Governor of California, 2026). The Information Commissioner’s chief executive and the German data protection conference reason from the blind bystander’s side: transparency mechanisms “assumed that people could see them,” and “Eine „Hinweis-LED” genügt hierfür nicht” (Information Commissioner’s Office, 2026; Datenschutzkonferenz, 2026). Zhao and colleagues record blind bystanders’ own demand, “If sighted people know, I should be able to know” (Zhao et al., 2023). Akter and colleagues found that both visually impaired and sighted respondents fear misrepresentation by camera-based assistance (Akter et al., 2020). The defect of October 2026 completes the case: a wearer who did nothing wrong can still record, and the credential and the indicator both fail that case (TechRadar, 2026; TechSpot, 2026).

8.7 Against a disability-specific path at all

The universal design camp turns the author’s own rule against the credential. CAST, the EALA and New America paper, the QAA, and EDRi argue that AI access should be designed for all. Adapting assessment is “preferable to banning,” and AI “should support, not replace, the individualized planning process” (CAST, 2025; EALA & New America, 2026; Quality Assurance Agency for Higher Education, 2023; European Digital Rights, 2024). Springer Nature’s Green tier, “Assistive AI use (low risk),” with “The goal is not to identify or penalise AI use,” and SAGE’s function-based rule that assistive tools “do not require disclosure,” are the objection in publisher form (Springer Nature, 2026; Sage Publishing, n.d.). A function-based rule for everyone needs no card from anyone. Staiman argues that “Style, fluency, or perceived ‘AI-ness’ of prose should be explicitly excluded from integrity judgments” for every author (Staiman, 2026). BaHammam’s 1.7 percent disclosure rate, with fear of bias as the cause, shows that the disabled writer’s fear is every writer’s fear (BaHammam, 2025). Fairplay’s model moratorium already carves out use under an IEP, a Section 504 plan, ADA Title II, or a provider’s written statement. A new credential would compete with an existing gate rather than fill a vacuum (Fairplay, 2026). The CRPD Committee places assistive technology on the accessibility side, owed to groups without request (Committee on the Rights of Persons with Disabilities, 2014; 2018). The RNIB places a proportionality duty on the venue rather than a right in the wearer’s hand (RNIB, 2026). Spain’s two provisions and H.R. 10515 offer a rights path and a vocational rehabilitation path, neither of which needs a card (BOE, 2026; H.R. 10515, 2026). The European Disability Forum’s own call is “to fully enforce the EU Artificial Intelligence Act,” not to issue a card (European Disability Forum, 2026b). France’s national disability conference of September 2026 set the goal of moving from an inclusive society “à une société pleinement accessible” (Conférence Nationale du Handicap, 2026). Against all of this stands the practitioner’s answer in Gose: “nothing in my closet says ‘one size fits all.'” Universal design “will help” but will not “make accommodations unnecessary” (Gose, 2016). The paper carries both.

8.8 The Union’s structural objection

Directive 2024/2841 confirms that the Union has no competence over disability assessment, excludes employment, social security, and long-term inclusion, and covers short stays only (Directive (EU) 2024/2841). Ireland’s National Disability Authority notes the card “does not affect a Member States’ competence,” and the European Parliament’s committee recorded that the directive “does not guarantee the mutual recognition of disability status” (National Disability Authority, n.d.; European Parliament, 2025). There is no single European agency to borrow. The European Union of the Deaf, EDRi, Autism-Europe, and Access Now hold that the AI Act names disabled people as at risk without “concrete actions,” that its bans “contain significant and disappointing loopholes,” and that inference of disability is “AI physiognomy” that “must be banned in all circumstances” (European Union of the Deaf, 2024; European Digital Rights, 2024; Autism-Europe, 2024; Access Now & European Digital Rights, 2023). The European Disability Forum warns that the 2026 Omnibus expands processing of disability data without consent for bias detection (European Disability Forum, 2026c). The floor the European credential would rest on is contested by the people it is meant to protect.

9. Questions for the Community and the Public

Each question arises where the source systems conflict or leave the point open. The paper asks them and does not settle them. The first 22 were set on October 13, 2026 and are carried as set, with question 5 reworded into a question on November 6, 2026. Questions 23 to 25 arise from the facts verified since, question 26 from the review of October 30, 2026, question 27 from the rulings of November 6, 2026, and question 28 from the review of November 7, 2026. A reading guide, which leaves the numbering and the wording as set: eligibility and certification are questions 1, 2, 3, 10, 12, and 16; the validation and its records are 4, 5, 6, 7, 8, 11, 25, and 26; institutions and platforms are 9, 13, 20, 23, and 28; benefits are 14, 15, and 17; disclosure is 27; devices are 18, 19, 21, 22, and 24.

1. Threshold. Where below unemployability should the line sit on SSA’s five-point scale?
2. Additional automatic qualifiers. Which of these should qualify automatically, beyond SSDI: SSI, which uses the same adult definition and needs no work record; an IEP or Section 504 plan for students; a VA disability rating; an accommodation already approved by an employer or school; eligibility for vocational rehabilitation services; a state disability card or identification marker? Section 4.2 states what each shows and what it leaves open, and the paper is open to any of them.
3. Who certifies. Licensed physicians, as in ABLE and parking? The wider set of medical sources SSA accepts for mental impairments, including psychologists, psychiatric nurse practitioners, licensed clinical social workers, and clinical mental health counselors? Or self-attestation alone? How does any answer square with the first companion paper’s line that authority over one’s own assistive use rests on no credential, and with the service-animal rule, which gates on function and conduct and forbids a demand for documentation?
4. Who issues the credential. A government agency, a standards body, the institutions’ own accommodation offices, or the platforms jointly? And how is the credential itself made accessible to its holder, which the European Disability Forum reports the Union’s own card is not?
5. Where records live and who protects them. Where should records live, and who protects them? A central registry of disabled AI users is a target, and the Department of Labor sent accommodation requesters’ functional-limitation categories outside its network in 2026. If no registry should exist, what method of verification and revocation meets security, privacy, and accessibility at once, and who holds the diagnosis the issuer sees?
6. Temporary needs. Parking caps temporary placards at six months, and SSA requires 12 months. What tier serves recovery and episodic conditions?
7. Reciprocity and adoption. Who must honor the credential? Is adoption voluntary, as the parking guidelines are, or required, as the parking rule’s recognition clause is among the states?
8. Misuse. How does the system deter false claims without moving the burden back onto genuine users, given the placard record?
9. Platforms. How does a platform honor a private credential without revealing it in what readers see? Can platforms separate substitution from disability-mediated assistance without forcing users to disclose a disability?
10. Undiagnosed people. How does a system built on medical determination reach a person with a real functional barrier and no diagnosis, when more than half of adults with an ADHD diagnosis in the United States received it as adults?
11. Another interpretation. A credential system is itself an institutional determination. Reciprocity spares a person from disclosing repeatedly, but the first disclosure still falls on the person. Could the system become one more interpretation that follows a person?
12. Supported certification. People with executive function or communication difficulties may struggle with the procedures meant to grant help. SSA accepts evidence from family, caregivers, and case managers with the person’s permission. How does a supporter help a person certify and present without taking over the decision?
13. Acting on a reading. A detector score or a human impression is a reading, not a record. Should any anti-AI rule be allowed to penalize a person on a reading alone, without a named human deciding on evidence?
14. AI-enabled work and disability reviews. SSA can weigh medical evidence of recovery at any time. Should work done with AI as an accommodation be treated like the subsidies and special conditions SSA already counts, so that it is not read as recovery? Who carries that question to SSA?
15. General-purpose AI as a work expense. Impairment-related work expenses cover assistive technology specifically designed to accommodate the impairment and exclude software unrelated to the disability and the work. General-purpose AI used as an accommodation sits between the two. Should it qualify, and should the first paper’s functional test be the showing?
16. One credential or tiers. Should the credential differ across functional, partial, and complete, or stay one credential with the level held privately? A visible tier would reveal severity. Readers of the fourth draft also read the three ranges as mixing a benefit status, a degree, and a basis. They proposed three evidence pathways in their place: an administrative determination, a professional attestation, and a functional declaration. The paper carries that reading here rather than restructuring Section 4.1 ahead of the community.
17. Unpaid productive activity. The paper holds that a person should not be punished for a podcast, a blog, or artwork. SSA generally excludes hobbies from substantial gainful activity, yet counts the kind of work usually done for pay whether or not profit is realized, and allows work below the earnings line to show ability. Where does creative and expressive activity done with AI fall, and what form should its protection take: SSA guidance, an explicit category in the standard, or both?
18. One sensor, two uses. The camera that reads a menu for a legally blind wearer is the camera bystanders fear. How does a venue grant the accommodation and protect bystanders at the same time, and what weight should the Hamburg view carry, that a visible cane or guide dog makes the aid predictable to others, against a private credential?
19. Blanket device bans. Should every ban on recording devices carry the model clause’s accommodation path by default, as Polk County’s school rule does and as the New York courts’ memorandum and the cruise and venue bans do not?
20. Prior notice as a condition. Polk County lets students with disabilities use smart glasses only after contacting the school, and Spain requires accreditation of need for public selection processes. Is advance notice a fair accommodation process, or the first burden of question 11 in working form?
21. Device classification. If a vendor’s glasses are classified as an assistive device for Medicare and Medicaid, does that settle question 15 for one product while deepening dependence on one vendor?
22. Public hostility. A credential does not reach a stranger. What protects a person using a device as an accommodation from confrontation in public, and whose job is that?
23. A right or a procedure. Spain now gives the person a statutory right to use support products without forced deactivation, and protects the person without official recognition. Does a right in the person make a credential unnecessary, or does the right need the credential to be exercised against a platform or an employer that does not know the person holds it?
24. The defective device. When the device records without the wearer’s command, as the Ray-Ban Meta glasses did in October 2026, the wearer did nothing wrong and the bystander was still recorded. Who answers for that, and does any credential or indicator design address it?
25. Protection without a card. The CRPD Committee, the European Disability Forum, and Spain’s Article 64.2 each hold that protection should not depend on proving status. If the community rejects a credential, what instrument gives a platform or employer notice of the person’s means without one?
26. Misuse against minimum necessary. The validation’s prize is relief from an AI rule or a detector, which may be worth more to a person without a disability than a parking space is. Deterring false claims would require the recipient to probe the person’s use, and the notice’s minimum-necessary rule forbids that probe. Where should the burden of deterrence sit, and can it sit anywhere other than on the issuer at the point of certification?
27. Disclosure after the punishment ends. Once a rule no longer punishes AI use, does a user in the gray area owe anyone a disclosure that a user at the gate does not, and who decides? The scale is where fairness to the person labeled normal is argued. Is disclosure the right instrument for that argument, or is open access to the tool the whole answer?
28. Which register works. The paper models its letter on the cease and desist and offers a first-notice variant in Section 3.5. Which produces the faster and fairer response from an accommodation office: a request that presumes the recipient did not know, or a notice that records that it now does? Does the answer differ by recipient class?

10. Who the Paper Invites into the Questions

The paper invites answers to the questions in Section 9, and it invites something more. Each notice served under Section 3, with its dated response, is a record of how an institution met AI as a disability tool. Those who serve notice are asked to report what comes back, so that the shape of refusal, penalty, and compelled disclosure can be seen in the record rather than argued from cases. What comes back is reported on one rubric: the recipient class, the date sent and the channel, the register used, the response and its date, whether a named person decided, the outcome (granted, equally effective alternative, lawful denial with the reason stated, no response), and any adverse action within six months. The reports are aggregate. A report carries the rubric’s fields and nothing else: no copy of the notice, no name, no diagnosis, and no institution named where naming it would identify the sender. The author accepts no identifiable notices or case files, and will not until a collection protocol with consent, redaction, access, retention, and withdrawal rules is published separately. Participation is voluntary, the denominator is received eligible reports, and the reports are read as the experience of those who sent notices, not as the prevalence of the problem.

The disability community first. The American Foundation for the Blind’s guiding principles hold that “Students and employees with disabilities should be able to have access to AI as an assistive technology,” and that schools and employers should consult the disability community when developing policies on AI as a reasonable accommodation (American Foundation for the Blind, 2025). Its June 2026 employment study found that 68 percent of respondents used AI at work and that one worker’s accommodation was made “no longer valid” by a rule barring contact uploads. It recommends that employers “consider the use of AI as an assistive technology when developing workplace AI policies” (Shock & Silverman, 2026). The National Federation of the Blind, the American Council of the Blind, the Autistic Self Advocacy Network, the European Disability Forum, the RNIB, and the organizations deploying the glasses have each taken positions this paper carries in Section 8. Each is invited here to answer the questions in Section 9 in its own words.

Voices found while exploring, and more to find. Each is one example of writers whose style draws misreadings from detectors and from readers. Each is an illustration, not evidence. Eva Redford’s statement, “normal is not neutral, and it never was,” as carried on the Talk Architecture podcast of August 19, 2026 (Talk Architecture, 2026). Rua Williams, a Purdue University professor whose email a reader mistook for AI: “I’m not an AI. I’m just autistic” (The College Fix, 2023). Gegg-Harrison and Quarterman tested popular detectors on their own writing, found false positive rates well above the vendors’ claims, and discussed their own neurodivergence as a possible factor (as cited in Chambers & Kelley, 2025). The paper asks for more voices like these, especially people whose natural writing has been flagged or doubted.

Prior work to position against. Accessibility Standards Canada’s CAN-ASC-6.2:2025 requires, in clause 10.2.3, that organizations evaluate inequities “in collaboration with the intended beneficiary” when AI-based technology is considered for an accommodation. It contains no credential, notice, or portability mechanism (Accessibility Standards Canada, 2025). The Autistic Self Advocacy Network’s July 2025 position on generative AI in plain language (Autistic Self Advocacy Network, 2025). The W3C’s accessibility of machine learning and generative AI document, an Editor’s Draft of March 28, 2026, names in its scope “users applying ML-based AI themselves, including in assistive technologies.” It does not address authorship, policies restricting AI use, or accommodations (W3C, 2026). Caldeira’s proposal for assistive AI as a “cognitive right” for neurodivergent people in Brazil, framed as entitlement to provision rather than as a shield against anti-AI rules (Caldeira, 2025). Pierrès, Darvishy, and Christen’s finding that disabled students prefer general permission with clear guidelines to an accommodation they must disclose to obtain (Pierrès et al., 2025). Sidorkin’s report that some disability offices have added AI uses to accommodation letters, and his rule that for detectors “a flag is not proof” (Sidorkin, 2025). Zhuang and Goggin’s question, “how can we govern generative AI for disability and inclusion?” (Zhuang & Goggin, 2026). The paper is not first to argue that AI can be a reasonable accommodation. Sidorkin, Brophy, the National Deaf Center, the American Foundation for the Blind, and others argued it first (Brophy, 2025; National Deaf Center on Postsecondary Outcomes, 2025). What the paper adds is the instrument.

Operators of the systems the framework borrows from: the ABLE National Resource Center, state Protection and Advocacy agencies, Developmental Disabilities Councils, and the state agencies that issue placards under 23 C.F.R. Part 1235.

People who already help beneficiaries work: benefits specialists and vocational rehabilitation agencies, which SSA names in its work incentive rules, bear directly on questions 14, 15, and 17. H.R. 10515 would put the question to every state vocational rehabilitation plan.

People and institutions inside the glasses case: blind and low-vision wearers, the organizations deploying the glasses, the bystander groups now in court in Paris and Amsterdam, the data protection authorities on both sides of the Atlantic, and the venues, courts, schools, and carriers writing device rules, on questions 18 through 25.

Lawmakers in Spain and the ministry drafting the reasonable adjustments framework due by October 23, 2027, who will set how need is accredited and may say whether the dependency law’s definition of a support product governs Article 7.6. They may find in the first companion paper’s functional test a definition the law does not yet contain.

11. Conclusion

A person who uses AI as a disability tool is punished three ways today: by refusal, by a machine’s reading, and by a label that costs them readers. The protections against all three already exist, where each protection’s trigger is met. The ADA defines disability by function and makes failure to accommodate a form of discrimination. It keeps the accommodation in a separate file and forbids the employer to disclose it. Title II tells a public entity to give primary consideration to the person’s request for an auxiliary aid and to modify its rules when modification is necessary. Section 504 reaches every funded program. The GDPR makes disability data a special category for every controller it reaches, and Directive 2000/78 requires appropriate measures where needed. Spain has now written the right to benefit from AI, and the right to use a support product without being made to switch it off, into statute. What no law gives the person is a way to say, once and in a form the next institution recognizes, that this tool is that person’s means, and that the institution now knows it.

That is the job of a cease and desist, and it is the job of this paper’s notice: to name the protection, name the conduct, state the standing, and make the recipient knowing. Behind it, a validation built from the Social Security definition, the SSDI award, the ABLE certification, the parking placard’s face, and the TSA card’s discretion lets the recipient believe the standing without opening a medical file. Where the law reaches the recipient, the duty to consider the notice attaches on arrival. Where it does not, the notice builds the record that every later remedy needs, and the paper claims nothing more for it. The notice is written to teach as much as to warn, because the institutions that punish this use mostly don’t know what they are punishing. Each notice served, with its answer, is one record of how far that punishment reaches. The call is to serve notice, keep the record, and report what comes back.

The paper proposes the instrument, states what it would and would not do, carries the full case against it, and asks 28 questions that the author cannot answer alone. The community that would hold the validation should design it, or reject it and say what should stand in its place. The one position the paper holds as its own is the position of the series: the same standard, a different means, and no punishment for the attempt.

References

Works by the author, with the sources supporting each

Each entry leads with the work’s first public record and nests the version cited. The indented lines are that work’s own sources, listed so that a self-citation can be checked without opening the companion, and some repeat entries in the Sources list for that reason.

• Puglisi, B. C. (n.d.). Checkpoint-Based Governance [Canonical framework page]. basilpuglisi.com. https://basilpuglisi.com/cbg/
  • Asimov, I. (1942). Runaround. Astounding Science Fiction.
  • Asimov, I. (1985). Robots and Empire. Doubleday.
• Puglisi, B. C. (2025a, April 28). Building authority with verified AI research [Two versions, #AIa Originality.ai review]. basilpuglisi.com. https://basilpuglisi.com/building-authority-with-verified-ai-research-two-versions-aia-originality-ai-review/
  • Acrolinx. (2025, March 5). AI and the law: Navigating legal risks in content creation. https://www.acrolinx.com/blog/ai-laws-for-content-creation
  • Anthropic. (2024, March 4). Introducing the next generation of Claude. https://www.anthropic.com/news/claude-3-family
  • AWS News Blog. (2024, March 27). Anthropic’s Claude 3 Sonnet model is now available on Amazon Bedrock. https://aws.amazon.com/blogs/aws/anthropic-claude-3-sonnet-model-is-now-available-on-amazon-bedrock/
  • ElevenLabs. (2025, March 17). March 17, 2025 changelog. https://elevenlabs.io/docs/changelog/2025/3/17
  • FusionForce Media. (2025, February 25). Perplexity AI: Master content creation like a pro in 2025. https://fusionforcemedia.com/perplexity-ai-2025/
  • Google Cloud. (2024, March 14). Anthropic’s Claude 3 models now available on Vertex AI. https://cloud.google.com/blog/products/ai-machine-learning/claude-3-on-vertex-ai
  • Harvard Business School. (2025, March 31). Perplexity: Redefining search. https://www.hbs.edu/faculty/Pages/item.aspx?num=67198
  • Influencer Marketing Hub. (2024, December 1). Perplexity AI SEO: Is this the future of search? https://influencermarketinghub.com/perplexity-ai-seo/
  • Inside Privacy. (2024, March 18). China releases new labeling requirements for AI-generated content. Covington & Burling LLP. https://www.insideprivacy.com/international/china/china-releases-new-labeling-requirements-for-ai-generated-content/
  • McKinsey & Company. (2025, March 12). The state of AI: Global survey. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
  • Perplexity. (2025, January 4). Answering your questions about Perplexity and our partnership with AnyDesktop. https://blog.perplexity.ai/anydesktop
  • Perplexity. (2025, February 13). Introducing Perplexity Enterprise Pro. https://blog.perplexity.ai/enterprise-pro
  • Quora. (2024, March 5). Poe introduces the new Claude 3 models, available now. https://quorablog.quora.com/Poe-introduces-the-new-Claude-3-models-available-now
  • Solveo. (2025, March 3). 7 AI tools to dominate podcasting trends in 2025. https://www.solveo.co/post/7-ai-tools-to-dominate-podcasting-trends-in-2025
  • Surfer SEO. (2025, January 27). What’s new at Surfer? Product updates January 2025. https://surferseo.com/blog/january-2025-update/
  • YouTube. (2025, March 26). Descript March 2025 changelog: Smart transitions & Rooms improvements. https://www.youtube.com/watch?v=cdVY7wTZAIE
• Puglisi, B. C. (2025b, May 22). Why AI detection tools fail at measuring value [Opinion]. basilpuglisi.com. https://basilpuglisi.com/why-ai-detection-tools-fail-at-measuring-value-opinion/
  • Originality.ai. (2024, May). Robust AI Detection Study (RAID). https://originality.ai/blog/robust-ai-detection-study-raid
  • Fritz.ai. (2024, March 8). Originality AI: My honest review 2024. https://fritz.ai/originality-ai-review/
  • Scribbr. (2024, June 10). Originality.ai review. https://www.scribbr.com/ai-tools/originality-ai-review/
  • Software Oasis. (2023, November 21). Originality.ai review: Future of content authentication? https://softwareoasis.com/originality-ai-review-future-of-content-authentication/
  • Batt, R. (2023, May 5). The dark side of Originality.ai’s false positives. https://richardbatt.co.uk/2023/05/05/the-dark-side-of-originality-ais-false-positives-are-content-creators-at-risk/
  • Advanced Science News. (2023, July 12). AI detectors have a bias against non-native English speakers. https://www.advancedsciencenews.com/ai-detectors-have-a-bias-against-non-native-english-speakers/
  • Vanderbilt University. (2023, August 16). Guidance on AI detection and why we’re disabling Turnitin’s AI detector. https://www.vanderbilt.edu/brightspace/2023/08/16/guidance-on-ai-detection-and-why-were-disabling-turnitins-ai-detector/
  • Issues in Information Systems. (2024, March). Can GPTZero detect if students are using artificial intelligence? https://www.iacis.org/iis/2024/3_iis_2024_165-174.pdf
  • Gold Penguin. (2024, September 18). Writer.com AI detection tool review: Don’t even bother. https://goldpenguin.org/blog/writer-com-ai-detection-tool-review-dont-even-bother/
  • Capterra. (2025). Copyscape reviews 2025. https://www.capterra.com/p/179608/Copyscape/reviews/
• Puglisi, B. C. (2026a, May 31). Did AI write Magnifica Humanitas? Pope Leo XIV was the author, but what was the governance method? basilpuglisi.com. https://basilpuglisi.com/did-ai-write-magnifica-humanitas/
  • Pope Leo XIV. (2026, May 25). Magnifica humanitas: On safeguarding the human person in the time of artificial intelligence [Encyclical letter]. Vatican. https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html
  • Saha, S., & Feizi, S. (2025). Almost AI, almost human: The challenge of detecting AI-polished writing. University of Maryland. https://arxiv.org/abs/2502.15666
  • Jabarian, B., & Imas, A. (2025). Artificial writing and automated detection (Becker Friedman Institute Working Paper No. 2025-116). University of Chicago Booth School of Business. https://bfi.uchicago.edu/wp-content/uploads/2025/10/Artificial-Writing-and-Automated-Detection.pdf
  • Hadra, M., Cambridge, K., & Mesbah, M. (2026). Evaluating the accuracy and reliability of AI content detectors in academic contexts. International Journal for Educational Integrity, 22(4). https://doi.org/10.1007/s40979-026-00213-1
  • Garland, N. A. (2026). AI detectors fail diverse student populations: A mathematical framing of structural detection limits. Griffith University. https://arxiv.org/abs/2603.20254
  • Weber-Wulff, D., et al. (2023). Testing of detection tools for AI-generated text. International Journal for Educational Integrity, 19(26). https://doi.org/10.1007/s40979-023-00146-z
  • Liang, W., et al. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7). https://doi.org/10.1016/j.patter.2023.100779
  • Elkhatat, A., Elsaid, K., & Almeer, S. (2023). Evaluating the efficacy of AI content detection tools in differentiating between human and AI-generated text. International Journal for Educational Integrity, 19(17). https://doi.org/10.1007/s40979-023-00140-5
• Puglisi, B. C. (2026b, September 23). When AI becomes assistive technology: Artificially Intelligent Disability Assistance (AIDA) [Working paper]. Zenodo. https://doi.org/10.5281/zenodo.22908271
  • Version cited: https://doi.org/10.5281/zenodo.22908272
  • Adnin, R., & Das, M. (2024). “I look at it as the king of knowledge”: How blind people use and understand generative AI tools. In Proceedings of the 26th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS ’24). https://doi.org/10.1145/3663548.3675631
  • Accessibility Standards Canada. (2025). CAN-ASC-6.2:2025, Accessible and equitable artificial intelligence systems. https://accessible.canada.ca/creating-accessibility-standards/asc-62-accessible-equitable-artificial-intelligence-systems
  • American Foundation for the Blind. (2025, January). Guiding principles for more disability-inclusive AI. https://afb.org/research-and-initiatives/empowering-or-excluding/guiding-principles-more-disability-inclusive-ai
  • Autistic Self Advocacy Network. (2025, July). ASAN says no generative AI in plain language. https://autisticadvocacy.org/2025/07/asan-says-no-generative-ai-in-plain-language
  • Bickenbach, J. (2014). The capability approach and the International Classification of Functioning. ALTER: European Journal of Disability Research, 8, 10 to 23.
  • Brand, D., Fischer Mogensen, K., & Engelbrecht, M. (2026, September 15). AI as reasonable accommodation and disability employment. Disability & Society. https://doi.org/10.1080/09687599.2026.2726868
  • Burchardt, T. (2004). Capabilities and disability: The capabilities framework and the social model of disability. Disability & Society, 19(7), 735 to 751.
  • Coffey, K. V., Krahn, G. L., Hanley, J. P., & Neely, J. E. (2026). Identifying implicit bias in LLM-based chat AI toward people with intellectual disabilities. Disability and Health Journal.
  • CORDIS. (n.d.). AIDE: Adaptive multimodal interfaces to assist disabled people in daily activities (Horizon 2020 project 645322). European Commission. https://cordis.europa.eu/project/id/645322
  • Della Penna, G., Buzzi, M., & Leporini, B. (2026). Generative AI as a new assistive technology for web interaction. In Human-Computer Interaction, INTERACT 2025 (Lecture Notes in Computer Science, Vol. 16108). Springer. https://doi.org/10.1007/978-3-032-04999-5_6
  • Davis Wright Tremaine LLP. (2026, September). New technical standard for European Accessibility Act compliance released. https://www.dwt.com/insights/2026/09/european-accessibility-act-ict-standards-update
  • Demers, L., Weiss-Lambrou, R., & Ska, B. (2002). The Quebec User Evaluation of Satisfaction with Assistive Technology (QUEST 2.0). Technology and Disability, 14, 101 to 105.
  • Deque Systems. (2026, September). EN 301 549 v4.1.1 is final: What changed, what it means, and what you should do. https://www.deque.com/blog/en-301-549-v4-1-1-is-final-what-changed-what-it-means-and-what-you-should-do/
  • DeRosier, R., & Farber, R. (2005). Speech recognition software as an assistive device: A pilot study of user satisfaction and psychosocial impact. Work, 25(2), 125 to 134.
  • Eberhardt, K. E. (2026, July 7). Position paper: AI as a disability accommodation in higher education. Legal analysis and annotated bibliography on cognitive scaffolding and generative AI. Zenodo. https://doi.org/10.5281/zenodo.21231449
  • Ede, R. (2026, August 20). A thinking partner: Copilot as assistive technology. Defra digital, data, technology and security blog. https://defradigital.blog.gov.uk/2026/08/20/a-thinking-partner-copilot-as-assistive-technology/
  • El Morr, C., Kundi, B., Mobeen, F., Taleghani, S., El-Lahib, Y., & Gorman, R. (2024). AI and disability: A systematic scoping review. Health Informatics Journal, 30(3).
  • ETSI. (2026). EN 301 549 V4.1.1 (2026-09), Accessibility requirements for ICT products and services. https://www.etsi.org/deliver/etsi_en/301500_301599/301549/04.01.01_60/en_301549v040101p.pdf
  • Gilly, T. (2026a, August). Accommodation repricing: Disability, generative AI subscriptions, and the accessibility tax as a variable the vendor controls (v0.7). Real Safety AI Foundation. SSRN 7218538. https://doi.org/10.2139/ssrn.7218538
  • Gilly, T. (2026b). The accommodation axis: Disability discrimination, algorithmic opacity, and the deployers that anti-discrimination scholarship overlooks. SSRN 7011798. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7011798
  • Giraldo, M., & Sacchi, F. (2026). From technical performance to assistive validity: A critical review of evaluation approaches for AI-enabled assistive technology. Healthcare, 14(18), 2900. https://doi.org/10.3390/healthcare14182900
  • Glazko, K., Yamagami, M., Desai, A., Mack, K. A., et al. (2023). Three-month autoethnography of generative AI use for accessibility (arXiv:2308.09924). https://arxiv.org/abs/2308.09924
  • Gonzalez Penuela, R. E., Jung, C., Lin, S., Hu, R., & Azenkot, S. (2026). How multimodal large language models support access to visual information: A diary study with blind and low vision people. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3772318.3793266
  • Hadar Souval, D., Haber, Y., Tal, A., Simon, T., Elyoseph, T., & Elyoseph, Z. (2025). Transforming perceptions: Exploring the multifaceted potential of generative AI for people with cognitive disabilities. JMIR Neurotechnology, 4, e64182. https://doi.org/10.2196/64182
  • Hanuschock, W. E., & Silverman, A. M. (2026, August). Innovation for access: AI and its use as assistive technology for people with disabilities. American Foundation for the Blind. https://afb.org/research-and-initiatives/ai-series/innovation-access
  • Jang, J., Moharana, S., & Carrington, P. (2024). “It’s the only thing I can trust”: Envisioning large language model use by autistic workers for communication assistance. In Proceedings of the CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3613904.3642894
  • Johnson, J., Lewis, A., Mankoff, J., & Banner, O. (2026). “I don’t trust it, but I use it”: Navigating trust, privacy, and identity in disabled people’s use of generative AI. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3772318.3790652
  • Jutai, J., & Day, H. (2002). Psychosocial Impact of Assistive Devices Scale (PIADS). Technology and Disability, 14, 107 to 111.
  • Knobel, N. (2026, April 10). Algorithmic risk and neuroinclusion: AI, psychosocial safety, and the rights of neurodivergent workers in Aotearoa New Zealand. New Zealand Journal of Health and Safety Practice, 3(1). https://doi.org/10.26686/nzjhsp.v3i1.10606
  • Mitra, S. (2006). The capability approach and disability. Journal of Disability Policy Studies, 16(4), 236 to 247.
  • Mladenov, T. (2025). AI and disabled people’s independent living: A framework for analysis. AI & Society. https://doi.org/10.1007/s00146-025-02642-x
  • Moore, R. E., & Williams, A. B. (2020). AIDA: Using social scaffolding to assist workers with intellectual and developmental disabilities. In Companion of the 2020 ACM/IEEE International Conference on Human-Robot Interaction (pp. 366 to 368). https://doi.org/10.1145/3371382.3378385
  • Moore, R. E., & Williams, A. B. (2021). Towards a learning architecture to support social scaffolding for an artificially intelligent disability assistant. In Companion of the 2021 ACM/IEEE International Conference on Human-Robot Interaction (pp. 467 to 469). https://doi.org/10.1145/3434074.3447215
  • Morgan, D. (2026). Reasonable adjustments for neurodivergent students in higher education: Generative AI, accessibility, and mediated authorship. Disability & Society. https://doi.org/10.1080/09687599.2026.2667528
  • National Disability Insurance Agency. (n.d.). Framework for artificial intelligence-enabled assistive technology as supports under the NDIS. https://ndis.gov.au/news/8492-framework-artificial-intelligence-enabled-assistive-technology-supports-under-ndis
  • Nussbaum, M. C. (2006). Frontiers of justice: Disability, nationality, species membership. Harvard University Press. https://doi.org/10.4159/9780674041578
  • The Open University. (2025, July 15). Generative AI for students. https://about.open.ac.uk/policies-and-reports/policies-and-statements/generative-ai-learning-teaching-and-assessment-ou-0
  • Randall, K. N., Drew, H., Gilman, E. S., & Dixon, E. (2025). Assistive technology uses and barriers in the home and workplace for adults with intellectual and developmental disabilities. Journal of Applied Research in Intellectual Disabilities, 38(1), e13306. https://doi.org/10.1111/jar.13306 (first published online October 27, 2024)
  • Phillips, B., & Zhao, H. (1993). Predictors of assistive technology abandonment. Assistive Technology, 5(1), 36 to 45. https://doi.org/10.1080/10400435.1993.10132205
  • Polgar, J. M., Encarnação, P., Smith, E., & Cook, A. M. (2025). Assistive technologies: Principles and practice (6th ed.). Elsevier.
  • Rodgers, R. (n.d.). AI can create plain-language and easy-read versions of every written material on earth. Impact, 38(3). Institute on Community Integration, University of Minnesota. https://publications.ici.umn.edu/impact/38-3/ai-can-create-plain-language-and-easy-read-versions-of-every-written-material-on-earth-by-rylin-rodgers
  • Sustainable Bus. (2025, October). Masats AIDA: Artificial intelligence bus doors. https://www.sustainable-bus.com/components/masats-aida-artificial-intelligence-bus-doors
  • Scherer, M. J., & Craddock, G. (2002). Matching Person and Technology (MPT) assessment process. Technology and Disability, 14(3), 125 to 131.
  • Shew, A. (2023). Against technoableism: Rethinking who needs improvement. W. W. Norton.
  • Shock, A., & Silverman, A. M. (2026, June). Working with the machine: AI’s expanding role in employment for people with disabilities. American Foundation for the Blind. https://afb.org/research-and-initiatives/ai-series/working-machine
  • Silverman, A. M., Whistler, A. L., Shock, A., Heydarian, C. H., Baguhn, S. J., Hanuschock, W. E., Hashimoto, M., Khan, O., & Vader, M.-L. (2026). The AI quagmire: Benefits, risks, and user aspirations through a disability lens. American Foundation for the Blind. https://afb.org/AIResearch2
  • United Nations. (2006). Convention on the Rights of Persons with Disabilities. https://www.un.org/disabilities/documents/convention/convoptprot-e.pdf
  • Silvera-Tawil, D., Higgins, L., Packer, K., Bayor, A. A., Walker, J. G., Li, J., Niven, P., Khanna, S., Byrnes, J., Bradford, D., et al. (2025). AI-enabled AT framework: A principles-based approach to emerging assistive technology. Disability and Rehabilitation: Assistive Technology, 20, 1955 to 1974. https://doi.org/10.1080/17483107.2025.2479838
  • Valencia, S., Cave, R., Kallarackal, K., Seaver, K., Terry, M., & Kane, S. K. (2023). “The less I type, the better”: How AI language models can enhance or impede communication for AAC users. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3544548.3581560
  • van Toorn, G., Scully, J. L., & Gendera, S. (2025). “This robot is dictating her next steps in life”: Disability justice and relational AI ethics. AI & Society, 40. https://doi.org/10.1007/s00146-025-02224-x
  • Whittaker, M., Alper, M., Bennett, C. L., Hendren, S., Kaziunas, L., Mills, M., Morris, M. R., Rankin, J., Rogers, E., Salas, M., & West, S. M. (2019, November). Disability, bias, and AI. AI Now Institute. https://ainowinstitute.org/wp-content/uploads/2023/04/disabilitybiasai-2019.pdf
  • Williams, K., & Breazeal, C. (2013). Reducing driver task load and promoting sociability through an Affective Intelligent Driving Agent (AIDA). In Human-Computer Interaction, INTERACT 2013 (pp. 619 to 626). Springer.
  • Williams, A. B., Williams, R. M., Moore, R. E., & McFarlane, M. (2019). AIDA: A social co-robot to uplift workers with intellectual and developmental disabilities. In 2019 14th ACM/IEEE International Conference on Human-Robot Interaction (pp. 584 to 585). https://doi.org/10.1109/HRI.2019.8673272
  • World Health Organization. (2001). International classification of functioning, disability and health. WHO.
  • World Health Organization & UNICEF. (2022). Global report on assistive technology. WHO.
  • World Wide Web Consortium. (2024). Web Content Accessibility Guidelines (WCAG) 2.2. https://www.w3.org/TR/WCAG22/
  • Wright, C. (2026). Transcription is not generation: Distinguishing non-generative AI tool use from academic misconduct in higher education assessment. International Journal for Educational Integrity, 22, 24. https://doi.org/10.1007/s40979-026-00234-w
  • Zhao, X., Cox, A., & Chen, X. (2025). The use of generative AI by students with disabilities in higher education. The Internet and Higher Education, 66, 101014. https://doi.org/10.1016/j.iheduc.2025.101014
  • Zhao, X., Chen, X., & Cox, A. (2026). Exploring the affordances of generative AI in academic writing for students with disabilities. Policy Futures in Education, 24(1), 59 to 81. https://doi.org/10.1177/14782103251395436
• Puglisi, B. C. (2026c, September 24). When AI policy becomes an accessibility barrier: ADA exposure from restrictions on Artificially Intelligent Disability Assistance (AIDA) [Working paper]. Zenodo. https://doi.org/10.5281/zenodo.22940224
  • Version cited: https://doi.org/10.5281/zenodo.22940225
  • Adnin, R., Pandkar, A., Yao, B., Wang, D., & Das, M. (2025). Examining student and teacher perspectives on undisclosed use of generative AI in academic work. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems.
  • American Educational Research Association, American Psychological Association, & National Council on Measurement in Education. (2014). Standards for educational and psychological testing. American Educational Research Association.
  • American Foundation for the Blind. (2025, January). Guiding principles for more disability-inclusive AI. https://afb.org/research-and-initiatives/empowering-or-excluding/guiding-principles-more-disability-inclusive-ai
  • Associated Press. (2026, September 14). Trump calls AI risks a “hoax,” says there is a “SICK conspiracy” against AI and data centers. https://www.local10.com/business/2026/09/14/trump-calls-ai-risks-a-hoax-says-there-is-a-sick-conspiracy-against-ai-and-data-centers/
  • Brand, D., Fischer Mogensen, K., & Engelbrecht, M. (2026, September 15). AI as reasonable accommodation and disability employment. Disability & Society. https://doi.org/10.1080/09687599.2026.2726868
  • Buolamwini, J. (2026, April). [Comment on a LinkedIn post by B. C. Puglisi concerning the evocative audit]. LinkedIn.
  • Cisco. (2024). Cisco 2024 data privacy benchmark study. Cisco Systems.
  • Cisco. (2026). Cisco 2026 data privacy benchmark study. https://www.cisco.com/c/dam/en_us/about/doing_business/trust-center/docs/cisco-privacy-benchmark-study-2026.pdf
  • Cooley LLP. (2025, February 21). Gone but not forgotten: Federal laws still apply despite guidance disappearance act. https://www.cooley.com/news/insight/2025/2025-02-21-gone-but-not-forgotten-federal-laws-still-apply-despite-guidance-disappearance-act
  • Crootof, R., Kaminski, M. E., & Price, W. N., II. (2023). Humans in the loop. Vanderbilt Law Review, 76(2), 429 to 510.
  • Eberhardt, K. E. (2026, July 7). Position paper: AI as a disability accommodation in higher education. Legal analysis and annotated bibliography on cognitive scaffolding and generative AI. Zenodo. https://doi.org/10.5281/zenodo.21231449
  • Ede, R. (2026, August 20). A thinking partner: Copilot as assistive technology. Defra digital, data, technology and security blog. https://defradigital.blog.gov.uk/2026/08/20/a-thinking-partner-copilot-as-assistive-technology/
  • Elish, M. C. (2019). Moral crumple zones: Cautionary tales in human-robot interaction. Engaging Science, Technology, and Society, 5, 40 to 60. https://doi.org/10.17351/ests2019.260
  • Elsevier. (2026, June). Generative AI policies for journals. https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals
  • European Commission. (2026, July 20). Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of Regulation (EU) 2024/1689 (C(2026) 5054 final, Annex). https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems
  • Gallup. (2026). Americans cool toward AI. https://news.gallup.com/poll/712751/americans-cool-toward.aspx
  • GCN. (2026, August). LinkedIn slop flag used by a million members. https://gcn.com/linkedin-slop-flag-used-million-members/21090/
  • Gilly, T. (2026a, August). Accommodation repricing: Disability, generative AI subscriptions, and the accessibility tax as a variable the vendor controls (v0.7). Real Safety AI Foundation. https://realsafetyai.org/documents/accommodation_repricing_v0_7.pdf
  • Gilly, T. (2026b). The accommodation axis: Disability discrimination, algorithmic opacity, and the deployers that anti-discrimination scholarship overlooks. SSRN 7011798. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7011798
  • Government Executive. (2026a, August). Labor looks to AI to tackle accommodation requests from disabled employees. https://www.govexec.com/technology/2026/08/labor-department-seeks-ai-triage-accommodation-requests-disabled-employees/415438/
  • Government Executive. (2026b, August). Data from Labor employees seeking disability accommodations impacted by “internal incident.” https://www.govexec.com/management/2026/08/data-labor-employees-seeking-disability-accommodations-impacted-internal-incident/415673/
  • Hanuschock, W. E., & Silverman, A. M. (2026, August). Innovation for access: AI and its use as assistive technology for people with disabilities. American Foundation for the Blind. https://afb.org/research-and-initiatives/ai-series/innovation-access
  • ILRU. (n.d.). Public entities: Reasonable modifications and fundamental alteration. https://www.ilru.org/publications/public-entities-reasonable-modifications-fundamental-alteration
  • International Committee of Medical Journal Editors. (n.d.). Use of AI by authors. https://icmje.org/recommendations/browse/artificial-intelligence/ai-use-by-authors.html
  • Johnson, J., Lewis, A., Mankoff, J., & Banner, O. (2026). “I don’t trust it, but I use it”: Navigating trust, privacy, and identity in disabled people’s use of generative AI. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3772318.3790652
  • Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns. https://doi.org/10.1016/j.patter.2023.100779
  • Medium. (n.d.). Artificial intelligence (AI) content policy. Medium Help Center. https://help.medium.com/hc/en-us/articles/22576852947223-Artificial-Intelligence-AI-content-policy
  • Morgan, D. (2026). Reasonable adjustments for neurodivergent students in higher education: Generative AI, accessibility, and mediated authorship. Disability & Society. https://doi.org/10.1080/09687599.2026.2667528
  • The Open University. (2025, July 15). Generative AI for students. https://about.open.ac.uk/policies-and-reports/policies-and-statements/generative-ai-learning-teaching-and-assessment-ou-0
  • Perkins Coie. (n.d.). Eleventh Circuit vacates ruling that websites are not public accommodations under ADA. https://legacy.perkinscoie.com/insights/blog/eleventh-circuit-vacates-ruling-websites-are-not-public-accommodations-under-ada
  • Pew Research Center. (2025, September 17). How Americans view AI and its impact on people and society. https://www.pewresearch.org/wp-content/uploads/sites/20/2025/09/PS_2025.9.15_AI-and-its-impact_report.pdf
  • Pew Research Center. (2026, March 12). Key findings about how Americans view artificial intelligence. https://www.pewresearch.org/short-reads/2026/03/12/key-findings-about-how-americans-view-artificial-intelligence/
  • Phillips, B., & Zhao, H. (1993). Predictors of assistive technology abandonment. Assistive Technology, 5(1), 36 to 45. https://doi.org/10.1080/10400435.1993.10132205
  • Plagiarism Today. (2024, April 11). Medium sets new policies on AI-generated writing. https://www.plagiarismtoday.com/2024/04/11/medium-sets-new-policies-on-ai-generated-writing/
  • Quinnipiac University Poll. (2026, March 30). https://poll.qu.edu/poll-release?releaseid=3955
  • Raj, M., Berg, J. M., & Seamans, R. (2026). The artificial intelligence disclosure penalty: Humans persistently devalue AI-generated creative writing. Journal of Experimental Psychology: General. https://doi.org/10.1037/xge0001889
  • Sage Publishing. (n.d.). Artificial intelligence policy. https://www.sagepub.com/journals/publication-ethics-policies/artificial-intelligence-policy
  • Schilke, O., & Reimann, M. (2025). The transparency dilemma: How AI disclosure erodes trust. Organizational Behavior and Human Decision Processes, 188, 104405. https://doi.org/10.1016/j.obhdp.2025.104405
  • Sheridan College. (n.d.). Turnitin AI writing detection. https://ltsa.sheridancollege.ca/digital-learning-support-hub/turnitin-ai-writing-detection/
  • Shock, A., & Silverman, A. M. (2026, June). Working with the machine: AI’s expanding role in employment for people with disabilities. American Foundation for the Blind. https://afb.org/research-and-initiatives/ai-series/working-machine
  • Trump, D. J. (2026, September 14). [Post]. Truth Social. As reported by the Associated Press (2026).
  • White & Case LLP. (2026, August 4). EU AI Omnibus enters into force, amending the AI Act. https://www.whitecase.com/insight-alert/eu-ai-omnibus-enters-force-amending-ai-act
  • Wright, C. (2026). Transcription is not generation: Distinguishing non-generative AI tool use from academic misconduct in higher education assessment. International Journal for Educational Integrity, 22, 24. https://doi.org/10.1007/s40979-026-00234-w

Sources

• ABLE National Resource Center. (2026, February). Am I ABLE-eligible? https://www.ablenrc.org/am-i-able-eligible-and-how-do-i-open-an-account/
• Aboulafia, A. (2026, August 12). For Meta’s smart glasses, access is an afterthought. Tech Policy Press. https://techpolicy.press/for-metas-smart-glasses-access-is-an-afterthought
• Aboulafia, A., & Claypool, H. (2025, March 11). Building a disability-inclusive AI ecosystem. Center for Democracy and Technology and American Association of People with Disabilities. https://cdt.org/wp-content/uploads/2025/03/2025-03-11-CDT-Building-A-Disability-Inclusive-AI-Ecosystem-report-final.pdf
• Access Now & European Digital Rights. (2023, January). Prohibiting biometric categorisation (updated statement). https://www.accessnow.org/wp-content/uploads/2023/01/Prohibiting-Biometric-Categorisation-updated-version-January-2023.docx.pdf
• Accessibility Standards Canada. (2025). CAN-ASC-6.2:2025, Accessible and equitable artificial intelligence systems. https://accessible.canada.ca/creating-accessibility-standards/asc-62-accessible-equitable-artificial-intelligence-systems
• AI Incident Database. (n.d.). Incident 849: AI detection tools allegedly misidentify neurodivergent and ESL students’ work as AI-generated in academic settings. https://incidentdatabase.ai/cite/849/
• AirCaps. (2026, May 17). Smart glasses in 2026 [Vendor page]. https://aircaps.com/blog/smart-glasses-in-2026
• Akter, T., Ahmed, T., Kapadia, A., & Swaminathan, M. (2020). Privacy considerations of the visually impaired with camera based assistive technologies: Misrepresentation, impropriety, and fairness. In Proceedings of the 22nd International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS ’20). https://doi.org/10.1145/3373625.3417003
• American Civil Liberties Union et al. (2025, May 12). Coalition letter to Secretary Kennedy on the NIH autism data platform. https://assets.aclu.org/live/uploads/2025/05/2025.05.12-NIH-Data-Platform-Coalition-Letter-v8-Updated.pdf
• American Council of the Blind. (n.d.-a). Guidelines for the use of text to speech (TTS). https://www.acb.org/guidelines-for-use-of-text-to-speech-tts
• American Council of the Blind. (n.d.-b). Meta AI glasses and ACB. https://www.acb.org/glasses
• American Foundation for the Blind. (2025, January). Guiding principles for more disability-inclusive AI. https://afb.org/sites/default/files/2025-02/Guiding%20Principles%20for%20More%20Inclusive%20AI.pdf
• American Foundation for the Blind. (2026, January 7). AFB AI employment research [Blog]. https://www.afb.org/blog/entry/afb-ai-employment-research
• Amnesty International UK. (2025, July 10). Too much technology, not enough empathy. https://www.amnesty.org.uk/latest/uk-dwps-unhealthy-obsession-ai-discriminates-against-people-disabilities/
• Arteveldehogeschool. (2026). Onderwijs- en examenreglement 2026-2027. https://ahscdn.be/sites/default/files/2026-06/onderwijs_en_examenreglement_gra_ba_banaba_202627_online.pdf
• Ashman, J. J., Santo, L., Peters, Z. J., Okeyode, T., & Gidali, D. (2025, December). Visits to health centers by adults with attention-deficit/hyperactivity disorder: United States, 2023 (NCHS Data Brief No. 543). National Center for Health Statistics. https://www.cdc.gov/nchs/products/databriefs/db543.htm
• Autism-Europe. (2024, October 17). Advocating for autism in AI regulation: Navigating Europe’s new AI Act. https://www.autismeurope.org/blog/2024/10/17/advocating-for-autism-in-ai-regulation-navigating-europes-new-ai-act/
• Autistic Self Advocacy Network. (2025, July 29). ASAN says no generative AI in plain language. https://autisticadvocacy.org/2025/07/asan-says-no-generative-ai-in-plain-language/
• BaHammam, A. S. (2025). The transparency paradox: Why researchers avoid disclosing AI assistance in scientific writing. Nature and Science of Sleep, 17, 2569 to 2574. https://doi.org/10.2147/NSS.S568375
• Barandiaran, X. E., & Pérez-Verdugo, M. (2025). Generative midtended cognition and artificial intelligence: Thinging with thinging things. Synthese (accepted); arXiv:2411.06812. https://arxiv.org/html/2411.06812v1
• Best, C. (2026, July 21). Against Claudefishing. Substack. https://post.substack.com/p/against-claudefishing
• BestColleges. (2023). AI on job applications gives unfair advantage, survey finds. https://www.bestcolleges.com/research/ai-on-job-applications-gives-unfair-advantage-survey/
• Bisignano, F. (2026, June 10). Written testimony, House Committee on Ways and Means, SSA fiscal year 2027 budget. https://waysandmeans.house.gov/wp-content/uploads/2026/06/SSA-Testimony-FY-2027-Budget.pdf
• Bitkom. (2025). Generative KI im Unternehmen: Rechtliche Fragen zum Einsatz generativer Künstlicher Intelligenz im Unternehmen. https://www.bitkom.org/sites/main/files/2024-02/Bitkom-Leitfaden-Generative-KI-im-Unternehmen.pdf
• BOE. (2022). Real Decreto 888/2022, de 18 de octubre (BOE-A-2022-17105). https://www.boe.es/eli/es/rd/2022/10/18/888
• BOE. (2026). Ley 4/2026, de 1 de octubre (BOE-A-2026-20528). Boletín Oficial del Estado, núm. 246, October 3, 2026. https://boe.es/boe/dias/2026/10/03/pdfs/BOE-A-2026-20528.pdf
• BOE. (n.d.). Real Decreto Legislativo 1/2013, de 29 de noviembre, texto consolidado (in force from October 23, 2026). https://www.boe.es/buscar/act.php?id=BOE-A-2013-12632
• Boutin-Cooper, A., & Puglisi, A. (2026). Disability, technoethics, and power: Exploring AI refusal. Unbound: A Journal of Digital Scholarship, 4(1). https://doi.org/10.12794/journals.ujds.v4i1.367
• Brophy, M. (2025, November 26). The case for AI as accommodation. Inside Higher Ed. https://www.insidehighered.com/opinion/views/2025/11/26/case-ai-accommodation-opinion
• Brown, L. X. Z. (2021, December 14). Protecting disabled people’s privacy is a civil rights issue [Remarks before the NTIA listening session]. Center for Democracy and Technology. https://cdt.org/insights/protecting-disabled-peoples-privacy-is-a-civil-rights-issue-lydia-x-z-browns-remarks-before-the-ntias-listening-session-on-civil-rights-privacy-and-data/
• Caldeira, M. A. M. (2025). Assistive artificial intelligence as a cognitive right: A proposal for neurodivergent equity in the 21st century. Research, Society and Development, 14(12), e80141250271. https://doi.org/10.33448/rsd-v14i12.50271
• California State Auditor. (2017, April). Department of Motor Vehicles: Administrative and statutory changes will improve its ability to detect and deter misuse of disabled person parking placards (Report 2016-121). https://information.auditor.ca.gov/reports/2016-121/index.html
• Capelouto, J. D. (2026, September 1). Meta disables cameras on thousands of tampered smart glasses. Semafor. https://www.semafor.com/article/09/01/2026/meta-disables-cameras-on-thousands-of-tampered-smart-glasses
• CAST. (2025, February 27). 5 ways AI and UDL work better together. https://www.cast.org/wp-content/uploads/2025/03/UDL-AI-20250227-A11Y.pdf
• CBS News. (2026, August 21). Woman covertly filmed by Meta AI smart glasses raises privacy and consent concerns: “I had no say.” https://www.cbsnews.com/news/meta-ai-smart-glasses-covert-filming-privacy/
• Center for Democracy and Technology. (2022, May 25). Press release: Shining a light on how surveillance tech discriminates against disabled people. https://cdt.org/insights/press-release-shining-a-light-on-how-surveillance-tech-discriminates-against-disabled-people/
• Center on Budget and Policy Priorities. (2025, February 28). Chart book: Social Security Disability Insurance. https://www.cbpp.org/research/social-security/chart-book-social-security-disability-insurance
• Cha, S., Jackson, R., Shu, Y., Branham, S., & van der Hoek, A. (2026). Blind and low-vision software professionals and generative AI. In Proceedings of the 48th International Conference on Software Engineering (ICSE 2026). arXiv:2512.24462. https://arxiv.org/pdf/2512.24462
• Chambers, S., & Kelley, M. C. (2025). The misclassification of autistic writing as AI-generated. In Artificial Intelligence in Education (AIED 2025), Lecture Notes in Computer Science, 15879, 89 to 103. Springer. https://doi.org/10.1007/978-3-031-98420-4_7 (arXiv version posted after the chapter: arXiv:2607.14729, July 16, 2026, https://arxiv.org/abs/2607.14729)
• CNIL. (2026, June 29). Smart glasses: The CNIL calls for vigilance. https://cnil.fr/en/smart-glasses-cnil-calls-vigilance
• College Board. (2026). Guidance for artificial intelligence tools and other services (AP exam policies). https://apcentral.collegeboard.org/exam-administration-ordering-scores/administering-exams/exam-policies/artificial-intelligence-tools
• Committee on Publication Ethics. (2023, February 13). Authorship and AI tools (position statement). https://publicationethics.org/guidance/cope-position/authorship-and-ai-tools
• Committee on the Rights of Persons with Disabilities. (2014). General comment No. 2 (2014), Article 9: Accessibility (CRPD/C/GC/2). https://digitallibrary.un.org/record/812025/files/CRPD_C_GC_2-EN.pdf
• Committee on the Rights of Persons with Disabilities. (2018). General comment No. 6 (2018) on equality and non-discrimination (CRPD/C/GC/6). https://digitallibrary.un.org/record/1626976/files/CRPD_C_GC_6-EN.pdf
• Conférence Nationale du Handicap. (2026, September 4). Conférence nationale du handicap 2026: Dossier de presse. https://handicap.gouv.fr/sites/handicap/files/2026-09/Dossier-presse-CNH-2026.pdf
• Congreso de los Diputados. (2025, July 24). Proyecto de Ley 121/000064. Boletín Oficial de las Cortes Generales, Serie A, Núm. 64-1. https://www.congreso.es/public_oficiales/L15/CONG/BOCG/A/BOCG-15-A-64-1.PDF
• Congreso de los Diputados. (2026, July 22). Aprobación por el Pleno. Boletín Oficial de las Cortes Generales, Serie A, Núm. 64-8. https://www.congreso.es/public_oficiales/L15/CONG/BOCG/A/BOCG-15-A-64-8.PDF
• Congressional Research Service. (2025, December 4). Social Security Disability Insurance and Supplemental Security Income (Report R48748). https://www.everycrsreport.com/reports/R48748.html
• CORRECTIV. (2026, September 10). Datenschutzbehörde sieht bei Meta Smart Glasses Risiko heimlicher Aufnahmen. https://correctiv.org/aktuelles/kuenstliche-intelligenz/2026/09/10/datenschutzbehoerde-sieht-bei-meta-smart-glasses-risiko-heimliche-aufnahmen/
• Data Protection Commission. (n.d.). Case study: Processing of occupational health data. Ireland. https://www.dataprotection.ie/en/dpc-guidance/case-studies/disclosure-unauthorised-disclosure/processing-occupational-health-data
• Datenschutzkonferenz. (2026, September 24). Pressemitteilung zur 3. Zwischenkonferenz 2026 (smart glasses; Hamburg report), reproduced by the Landesbeauftragte für den Datenschutz und die Informationsfreiheit Baden-Württemberg. https://www.baden-wuerttemberg.datenschutz.de/dsk-fitnessstudios-smart-glasses-sicherheitsgesetze/
• Degener, T. (2016). Disability in a human rights context. Laws, 5(3), 35. https://doi.org/10.3390/laws5030035
• Disability Belongs. (2026, August 25). When disability is treated as a credibility problem. https://www.disabilitybelongs.org/2026/08/disability-credibility-problem/
• Disability Exchange. (2026, May 1). DDS state disability determination services wait times 2026. https://disabilityexchange.org/blog/dds-state-disability-determination-services-wait-times-2026/
• Disability Rights UK. (2021, November 23). GMCDP legal challenge to the DWP fraud algorithm. https://www.disabilityrightsuk.org/node/587
• Dispatches from the Autism Wars. (2026, September 6). Is AI detection software biased against autistic writers? Substack. https://dispatchesfromtheautismwars.substack.com/p/is-ai-detection-software-biased-against
• DutchNews. (2026, October 13). Privacy campaigners to ask court to ban sale of Meta AI glasses. https://www.dutchnews.nl/2026/10/privacy-campaigners-to-ask-court-to-ban-sale-of-meta-ai-glasses/
• EALA & New America. (2026, April). Prioritizing students with disabilities in AI policy (Version 2). https://www.newamerica.org/wp-content/uploads/2026/05/Prioritizing-Students-with-Disabilities-in-AI-Policy-av.pdf
• Ede, R. (2026, August 20). A thinking partner: Copilot as assistive technology. Defra digital, data, technology and security blog. https://defradigital.blog.gov.uk/2026/08/20/a-thinking-partner-copilot-as-assistive-technology/
• Electronic Frontier Foundation. (2026, February 4). Smart AI policy means examining its real harms and benefits. https://www.eff.org/deeplinks/2026/02/smart-ai-policy-means-understanding-its-real-harms-and-benefits
• Elish, M. C. (2019). Moral crumple zones: Cautionary tales in human-robot interaction. Engaging Science, Technology, and Society, 5, 40 to 60. https://doi.org/10.17351/ests2019.260
• Elsevier. (2026). Generative AI policies for journals. https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals
• e-Enfance. (2026a). Meta smart glasses: Concerns raised over women being filmed without their knowledge. https://e-enfance.org/en/meta-3018-smart-glasses-concerns-raised-over-women-being-filmed-without-their-knowledge/
• e-Enfance. (2026b, October 6). Lunettes connectées : le phénomène prend de l’ampleur, la justice se saisit du sujet. https://e-enfance.org/lunettes-connectees-le-phenomene-prend-de-lampleur-la-justice-se-saisit-du-sujet/
• Equal Employment Opportunity Commission. (2002, October 17). Enforcement guidance on reasonable accommodation and undue hardship under the ADA. https://www.eeoc.gov/laws/guidance/enforcement-guidance-reasonable-accommodation-and-undue-hardship-under-ada
• European Commission. (n.d.). European Disability Card and European Parking Card for persons with disabilities. https://commission.europa.eu/strategy-and-policy/policies/justice-and-fundamental-rights/disability/european-disability-card-and-european-parking-card-persons-disabilities_en
• European Commission. (2026, July 20). Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of Regulation (EU) 2024/1689 (C(2026) 5054 final, Annex). https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems
• European Data Protection Board. (2026, October 13). Statement 1/2026 on smart glasses. https://www.edpb.europa.eu/system/files/2026-10/20261013_edpb_statement_smart-glasses_en.pdf
• European Digital Rights. (2024, April 3). EU AI Act fails to set gold standard for human rights. https://edri.org/our-work/eu-ai-act-fails-to-set-gold-standard-for-human-rights/
• European Disability Forum. (2022, May). Position on the European Disability Card. https://www.edf-feph.org/content/uploads/2022/05/EDF-Position-on-the-European-Disability-Card-2022_final.pdf
• European Disability Forum. (2023, April 1). Resolution on the EU Artificial Intelligence Act for the inclusion of persons with disabilities. https://www.edf-feph.org/content/uploads/2023/04/EDF-Board-Resolution-on-the-EU-Artificial-intelligence-Act-for-the-inclusion-of-persons-with-disabilities.pdf
• European Disability Forum. (2026a, August 19). Delegated act on the European Disability Card Directive adopted. https://www.edf-feph.org/delegated-act-on-the-european-disability-card-directive-adopted/
• European Disability Forum. (2026b, June 8). Building inclusive AI: Why persons with disabilities must shape Europe’s digital future. https://www.edf-feph.org/building-inclusive-ai-why-persons-with-disabilities-must-shape-europes-digital-future/
• European Disability Forum. (2026c, June 23). 6 ways the AI Omnibus weakens protections for persons with disabilities. https://www.edf-feph.org/6-ways-the-ai-omnibus-weakens-protections-for-persons-with-disabilities/
• European Parliament. (2025, October 29). Report on the EU strategy for the rights of persons with disabilities post-2024 (A10-0211/2025). https://www.europarl.europa.eu/doceo/document/A-10-2025-0211_EN.html
• European Union of the Deaf. (2024, March 18). Historic moment as the EU Parliament adopts the first ever piece of legislation on AI; however, it falls short on human rights protections. https://www.eud.eu/historic-moment-as-the-eu-parliament-adopts-the-first-ever-piece-of-legislation-on-ai-however-it-falls-short-on-human-rights-protections/
• Fairplay. (2026, July). Model moratorium on generative AI products in school. https://fairplayforkids.org/wp-content/uploads/2026/07/Model-Moratorium-on-Generative-AI-Products-in-School.pdf
• Fancourt, D., & Finn, S. (2019). What is the evidence on the role of the arts in improving health and well-being? A scoping review (Health Evidence Network synthesis report 67). WHO Regional Office for Europe. https://www.who.int/europe/publications/i/item/9789289054553
• Federal Register. (2025a, January 31). Executive Order 14179, Removing barriers to American leadership in artificial intelligence. 90 FR 8741. https://www.federalregister.gov/documents/2025/01/31/2025-02172/removing-barriers-to-american-leadership-in-artificial-intelligence
• Federal Register. (2025b, December 16). Executive Order 14365, Ensuring a national policy framework for artificial intelligence. 90 FR 58499. https://www.federalregister.gov/documents/2025/12/16/2025-23092/ensuring-a-national-policy-framework-for-artificial-intelligence
• Federal Register. (2026a, April 20). U.S. Department of Justice, Extension of compliance dates for nondiscrimination on the basis of disability; accessibility of web information and services of state and local government entities (interim final rule). 91 FR 20902. https://www.federalregister.gov/documents/2026/04/20/2026-07663/extension-of-compliance-dates-for-nondiscrimination-on-the-basis-of-disability-accessibility-of-web
• Federal Register. (2026b, May 11). U.S. Department of Health and Human Services, Extension of compliance dates for nondiscrimination on the basis of disability; accessibility of web content and mobile applications. 91 FR 25496. https://www.federalregister.gov/documents/2026/05/11/2026-09266/extension-of-compliance-dates-for-nondiscrimination-on-the-basis-of-disability-accessibility-of-web
• Federal Register. (2026c, June 5). Executive Order 14409, Promoting advanced artificial intelligence innovation and security. 91 FR 34565. https://www.federalregister.gov/documents/2026/06/05/2026-11415/promoting-advanced-artificial-intelligence-innovation-and-security
• Federal Register. (2026d, October 2). Executive Order 14434, Inaugurating the era of Super Intelligence. 91 FR 63129. https://www.federalregister.gov/documents/2026/10/02/2026-20321/inaugurating-the-era-of-super-intelligence
• Frankowska-Takhari, S. (2026, May 21). Guest post: When AI helps write research, what happens to lived experience? The Scholarly Kitchen. https://scholarlykitchen.sspnet.org/2026/05/21/guest-post-when-ai-helps-write-research-what-happens-to-lived-experience/
• Freie Universität Berlin. (n.d.). Institut für Englische Philologie: Richtlinie zur Verwendung von KI-Tools in Studien- und Prüfungsleistungen. https://www.geisteswissenschaften.fu-berlin.de/we06/lehre_und_studium/Studium/Hinweise-zu-Hausarbeiten-und-Pruefungsleistungen/index.html
• Garante per la protezione dei dati personali. (2026, October 14). Smart glasses, i Garanti europei: Servono maggiori tutele per la privacy (Doc-Web 10307908). https://www.garanteprivacy.it/garante/doc.jsp?ID=10307908
• Gilly, T. (2026). Accommodation repricing: Disability, generative AI subscriptions, and the accessibility tax as a variable the vendor controls (v0.7). Real Safety AI Foundation; SSRN 7218538. https://realsafetyai.org/documents/accommodation_repricing_v0_7.pdf
• Gose, B. (2016, September 18). Disability experts debate merits of universal design. The Chronicle of Higher Education. https://www.chronicle.com/article/disability-experts-debate-merits-of-universal-design
• Greenhut, S. (2017, May 10). Auditor spurs calls to combat disabled parking spot abuse. R Street Institute. https://www.rstreet.org/commentary/auditor-spurs-calls-to-combat-disabled-parking-spot-abuse/
• H.R. 10515, Artificial Intelligence Education and Workforce Readiness Act of 2026, 119th Cong. (2026, September 21) (introduced text). https://www.govinfo.gov/content/pkg/BILLS-119hr10515ih/html/BILLS-119hr10515ih.htm
• Hamraie, A., & Fritsch, K. (2019). Crip technoscience manifesto. Catalyst: Feminism, Theory, Technoscience, 5(1), 1 to 33. https://doi.org/10.28968/cftt.v5i1.29607
• Harstone, E. (2024, October 11). On AI and accessibility. Authors Publish. https://authorspublish.com/on-ai-and-accessibility
• Hawkins, B. (2025, April 24). Proposal to create federal autism registry stokes fear in disability advocates. The 74. https://www.the74million.org/article/proposal-to-create-federal-autism-registry-stokes-fear-in-disability-advocates/
• Heaney, D. (2026, June 24). Meta is giving free smart glasses to every blind US veteran. UploadVR. https://www.uploadvr.com/meta-is-giving-free-smart-glasses-to-every-blind-us-veteran/
• Heckman, J. (2026a, August 27). Data of Labor employees seeking disability accommodations impacted in internal incident. Government Executive. https://www.govexec.com/management/2026/08/data-labor-employees-seeking-disability-accommodations-impacted-internal-incident/415673/
• Heckman, J. (2026b, August 31). 3,150 Labor employees’ disability accommodations data leaked, internal memo says. Government Executive. https://www.govexec.com/management/2026/08/3150-labor-employees-disability-accommodations-data-leak-internal-memo/415738/
• HmbBfDI. (2026, September 10). Final report on the technical and data protection review of the Ray-Ban Meta AI Glasses. Der Hamburgische Beauftragte für Datenschutz und Informationsfreiheit. https://datenschutz-hamburg.de/fileadmin/user_upload/HmbBfDI/Datenschutz/Informationen/260910_HmbBfDI_Abschlussbericht_Ray_Ban_Meta_AI_Glasses_EN.pdf
• Information Commissioner’s Office. (2026, September 17). Enabling technology for good [Blog by Paul Arnold]. https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/09/enabling-technology-for-good/
• INPS. (2026, February 24). Riforma della disabilità: Al via la terza fase sperimentale. https://inps.it/it/it/inps-comunica/notizie/dettaglio-news-page.news.2026.02.riforma-della-disabilit-al-via-la-terza-fase-sperimentale.html
• Internal Revenue Service. (2025). Revenue Procedure 2025-32. https://www.irs.gov/pub/irs-drop/rp-25-32.pdf
• International Committee of Medical Journal Editors. (n.d.). Defining the role of authors and contributors. https://www.icmje.org/recommendations/browse/roles-and-responsibilities/defining-the-role-of-authors-and-contributors.html
• Jamshed, H., Naseem, M., Potluri, V., & Brewer, R. N. (2025). Rethinking productivity with GenAI: A neurodivergent students’ perspective. In Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2025). https://doi.org/10.1145/3663547.3746329
• Johnson, J., Lewis, A., Mankoff, J., & Banner, O. (2026). “I don’t trust it, but I use it”: Navigating trust, privacy, and identity in disabled people’s use of generative AI. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3772318.3790652
• Journal of Oral and Maxillofacial Surgery. (2026, September). Guide for authors. Elsevier. https://legacyfileshare.elsevier.com/promis_misc/joms-guide-for-authors%20(2).pdf
• Kelley, S. (2025, Fall). Meta glasses review. AFB AccessWorld. https://afb.org/aw/fall2025/meta-glasses-review
• King’s College London. (2026, June 16). ADHD may be under-diagnosed in adults across England. https://www.kcl.ac.uk/news/adhdmay-be-under-diagnosed-in-adults-across-england-1
• Kirk, C. P., & Givi, J. (2025). The AI-authorship effect: Understanding authenticity, moral disgust, and consumer responses to AI-generated marketing communications. Journal of Business Research, 186, 114984. https://doi.org/10.1016/j.jbusres.2024.114984
• Lamb, C. E., Brown, D. G., & Grossman, M. R. (2024). Precarity and solidarity: Preliminary results on a study of queer and disabled fiction writers’ experiences with generative AI. arXiv:2412.04575. https://arxiv.org/abs/2412.04575
• Legal Cheek. (2026, August 13). Courts ban Meta glasses. https://www.legalcheek.com/2026/08/courts-ban-meta-glasses/
• Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), 100779. https://doi.org/10.1016/j.patter.2023.100779
• Legal Information Institute. (n.d.). Cease and desist letter. In Wex. Cornell Law School. https://www.law.cornell.edu/wex/cease_and_desist_letter
• Lemkin Institute for Genocide Prevention. (2025, May 27). The proposed autism registry. https://www.against-genocide.org/post/lemkin-institute-the-proposed-autism-registry
• Luria, M., Scherer, M., Thakur, D., Aboulafia, A., Claypool, H., & Negrón, W. (2024, November). Screened out: The impact of digitized hiring assessments on disabled workers. Center for Democracy and Technology. https://cdt.org/wp-content/uploads/2024/11/2024-11-20-CDT-Research-AEDT-report-final.pdf
• Macfarlane, K. A. (2021). Disability without documentation. Fordham Law Review, 90(1), 59. https://ir.lawnet.fordham.edu/flr/vol90/iss1/2/
• Macfarlane, K. A. (2025, July). Disability documentation and disability discrimination. Human Rights Magazine. American Bar Association. https://www.americanbar.org/groups/crsj/resources/human-rights/2025-july/disability-documentation-discrimination/
• McCrossin, R. (2022). Finding the true number of females with autistic spectrum disorder by estimating the biases in initial recognition and clinical diagnosis. Children, 9(2), 272. https://doi.org/10.3390/children9020272
• Medium. (n.d.). Artificial intelligence (AI) content policy. Medium Help Center. https://help.medium.com/hc/en-us/articles/22576852947223-Artificial-Intelligence-AI-content-policy
• Meta. (2026, June 12). The future is for everyone: Free AI glasses for every blind veteran in America. https://about.fb.com/news/2026/06/free-ai-glasses-for-every-blind-veteran/
• Ministerio de Derechos Sociales, Consumo y Agenda 2030. (2026, September 16). El Congreso aprueba la gran reforma de las leyes de dependencia y discapacidad. https://www.dsca.gob.es/es/comunicacion/notas-prensa/congreso-aprueba-gran-reforma-dependencia-discapacidad-nuevos-derechos
• Mitchell, J., & Thompson, J. (2025, August). Race, ethnicity, and SSDI allowance rates (CES Working Paper 25-54). U.S. Census Bureau. https://cdn.www.census.gov/library/working-papers/2025/adrm/CES-WP-25-54.html
• Mosen, J. (2025, October). Convention address. Braille Monitor. National Federation of the Blind. https://nfb.org/images/nfb/publications/bm/bm25/bm2509/bm250905.htm
• National Council on Disability. (2025, October 31). 2025 progress report. https://www.ncd.gov/assets/uploads/reports/2025/ncd-2025-progress-report.pdf
• National Council on Disability. (2026, March 3). Statement for the record, Senate Subcommittee on Science, Manufacturing, and Competitiveness, regarding AI. https://www.ncd.gov/testimonies/2026-03-03-statement-for-the-record-senate-science-manufacturing-and-competitiveness-subcommittee-regarding-ai/
• National Deaf Center on Postsecondary Outcomes. (2025, September 17). AI as an accommodation for deaf college students. https://nationaldeafcenter.org/news-items/ai-as-an-accommodation-for-deaf-college-students/
• National Disability Authority. (n.d.). Frequently asked questions: European Disability Card. Ireland. https://nda.ie/publications/frequently-asked-questions-european-disability-card
• National Federation of the Blind. (2025). 2025 resolutions. https://nfb.org/resources/speeches-and-reports/resolutions/2025-resolutions
• National Federation of the Blind. (2026). 2026 resolutions (adopted July 7, 2026). https://nfb.org/resources/speeches-and-reports/resolutions/2026-resolutions
• New York State Unified Court System. (2026, July 1). Amended: Prohibition against smart glasses in court facilities [Memorandum, effective July 20, 2026]. https://webfiles.nycourts.gov/public/2026-07/amendedprohibitionagainstsmartglassesmemo.pdf
• Office of the Governor of California. (2026, September 30). Veto message, Senate Bill 1130. https://www.gov.ca.gov/wp-content/uploads/2026/09/VETO-msg-SB-1130.pdf
• Offlimits. (2026, October 15). Offlimits en The Firewall dagvaarden Meta [News]. https://offlimits.nl/
• Oldham, C. (2026, April 27). Disability, assistive technology and integrity double standards. International Center for Academic Integrity. https://academicintegrity.org/aws/ICAI/pt/sd/news_article/618492/_PARENT/layout_details/false
• Olive. (2026, September). EEOC pulled AI guidance. https://olive.is/answers/eeoc-pulled-ai-guidance/
• Palamountain, C. (2026, June 1). National Federation of the Blind challenges last-minute deadline extensions for website and mobile app accessibility. Seyfarth ADA Title III News and Insights. https://www.adatitleiii.com/2026/06/national-federation-of-the-blind-challenges-last-minute-deadline-extensions-for-website-and-mobile-app-accessibility/
• Pangram Labs. (2026, October 16). Is Pangram accurate? https://www.pangram.com/blog/is-pangram-accurate
• Pavlou, D. (2026, September 4). Smart glasses helped a man see the world. The privacy panic took it away. Straight Arrow News. https://san.com/cc/smart-glasses-gave-a-man-sight-the-privacy-panic-took-it-away/
• Pennsylvania House of Representatives. (2026, June 4). Rep. Ciresi news release on House Bill 2603. https://www.pahouse.com/Ciresi/InTheNews/NewsRelease/?id=143905
• Pierrès, O., Darvishy, A., & Christen, M. (2025). Exploring the role of generative AI in higher education: Semi-structured interviews with students with disabilities. Education and Information Technologies, 30, 8923 to 8952. https://doi.org/10.1007/s10639-024-13134-8
• Polk County Public Schools. (2026, August 13). Important notice regarding smart glasses. https://polkschoolsfl.com/news/important_notice_regarding_smart_glasses
• Putnam, D., Wasserman, D., Blustein, J., & Asch, A. (2026). Disability and justice. In Stanford Encyclopedia of Philosophy (first published May 23, 2013; substantive revision May 29, 2026; maintained by J. Begon). https://plato.stanford.edu/entries/disability-justice/
• Quality Assurance Agency for Higher Education. (2023, May 8). QAA publishes additional advice on generative artificial intelligence tools. https://www.qaa.ac.uk/news-events/news/qaa-publishes-additional-advice-on-generative-artificial-intelligence-tools
• Raj, M., Berg, J. M., & Seamans, R. (2026). The artificial intelligence disclosure penalty: Humans persistently devalue AI-generated creative writing. Journal of Experimental Psychology: General. https://doi.org/10.1037/xge0001889
• Reif, J. A., Larrick, R. P., & Soll, J. B. (2025). Evidence of a social evaluation penalty for using AI. Proceedings of the National Academy of Sciences, 122(19), e2426766122. https://doi.org/10.1073/pnas.2426766122
• RNIB. (2026, September 14). AI smart glasses as assistive technology. Royal National Institute of Blind People. https://www.rnib.org.uk/news/ai-smart-glasses-as-assistive-technology/
• Rojas, H. (2026, August 25). LinkedIn’s “slop” flag used by a million members. GCN. https://gcn.com/linkedin-slop-flag-used-million-members/21090/
• Royal Caribbean. (n.d.). Prohibited items onboard policy. https://www.royalcaribbean.com/faq/questions/prohibited-items-onboard-policy
• Sage Publishing. (n.d.). Artificial intelligence policy. https://www.sagepub.com/journals/publication-ethics-policies/artificial-intelligence-policy
• Schilke, O., & Reimann, M. (2025). The transparency dilemma: How AI disclosure erodes trust. Organizational Behavior and Human Decision Processes, 188, 104405. https://doi.org/10.1016/j.obhdp.2025.104405
• Security.org. (2026, August 13). 2026 smart glasses market report and adoption statistics. https://www.security.org/resources/smart-glasses-statistics-report/
• Service Public. (2026). Reconnaissance de la qualité de travailleur handicapé (RQTH) (F1650). https://www.service-public.gouv.fr/particuliers/vosdroits/F1650
• Seung, Y., & Basham, J. D. (2026). Cognitive offloading in the age of generative AI: What does it mean for students with learning disabilities? Learning Disability Quarterly, 49(3), 121 to 133. https://doi.org/10.1177/07319487261439132
• Shock, A., & Silverman, A. M. (2026, June). Working with the machine: AI’s expanding role in employment for people with disabilities. American Foundation for the Blind. https://afb.org/sites/default/files/2026-07/Working%20with%20the%20Machine%20-%20AI%20Employment%20Report%20-%20Accessible.pdf
• Sidorkin, A. M. (2025). AI as an academic accommodation for students with disabilities. AI and Education. https://doi.org/10.36851/ai-edu.vi.5282
• Silverman, A. M., et al. (2026, March). The AI quagmire: Benefits, risks, and aspirations through a disability lens. American Foundation for the Blind. https://afb.org/sites/default/files/2026-07/The%20AI%20Quagmire%20White%20Paper%20-%20Final%20Accessible.pdf
• Sins Invalid. (n.d.-a). 10 principles of disability justice [Blog]. https://sinsinvalid.org/blog/10-principles-of-disability-justice
• Sins Invalid. (n.d.-b). 10 principles of disability justice (2nd ed.) [PDF]. https://www.sinsinvalid.org/s/10_Principles_of_DJ-2ndEd.pdf
• Smart Glasses Australia. (2026, September 21). Protect smart glasses for people with disabilities [Petition]. Change.org. https://www.change.org/p/protect-smart-glasses-for-people-with-disabilities
• Social Security Administration. (n.d.-a). Disability evaluation under Social Security: Part I, general information. https://www.ssa.gov/disability/professionals/bluebook/general-info.htm
• Social Security Administration. (n.d.-b). 12.00 Mental disorders, adult. https://www.ssa.gov/disability/professionals/bluebook/12.00-MentalDisorders-Adult.htm
• Social Security Administration. (n.d.-c). How you qualify for disability benefits. https://www.ssa.gov/benefits/disability/qualify.html
• Social Security Administration. (1996, July 2). SSR 96-8p: Policy interpretation ruling, Titles II and XVI: Assessing residual functional capacity in initial claims. https://www.ssa.gov/OP_Home/rulings/di/01/SSR96-08-di-01.html
• Social Security Administration. (2024). Annual statistical report on the Social Security Disability Insurance program, 2024 (Section 4). https://www.ssa.gov/policy/docs/statcomps/di_asr/2024/sect04.html
• Social Security Administration. (2026a). Substantial gainful activity. https://www.ssa.gov/oact/cola/sga.html
• Social Security Administration. (2026b). Trial work period. https://www.ssa.gov/oact/cola/twp.html
• Social Security Administration. (2026c). Red Book: Returning to work; SSDI only employment supports. https://www.ssa.gov/redbook/eng/ssdi-only-employment-supports.htm
• Social Security Administration. (2026d). Red Book: SSDI and SSI employment supports. https://www.ssa.gov/redbook/eng/ssdi-and-ssi-employments-supports.htm
• Social Security Administration. (2026e). Spotlight on ABLE accounts, 2026 edition. https://www.ssa.gov/ssi/spotlights/spot-able.html
• Social Security Administration. (2026f). Disabled worker beneficiary statistics. https://www.ssa.gov/oact/STATS/dibStat.html
• Social Security Administration. (2026g, July 22). 2026 annual report of the Supplemental Security Income program. https://www.govinfo.gov/content/pkg/CMR-SSA1-00202586/pdf/CMR-SSA1-00202586.pdf
• Society for Human Resource Management. (2023, October 15). AI policies and risks. https://www.shrm.org/in/topics-tools/employment-law-compliance/ai-policies-and-risks
• Society for Human Resource Management. (2026, September 23). When workers break AI rules, employers may still pay the price. https://www.shrm.org/in/topics-tools/employment-law-compliance/when-workers-break-ai-rules-employers-may-still-pay-price
• Springer Nature. (2026). Editor AI use and assessment guidance. https://www.springernature.com/gp/policies/editorial-policies/ai-editor-guidance
• Staley, B. S., et al. (2024, October 10). Attention-deficit/hyperactivity disorder diagnosis, treatment, and telehealth use in adults: National Center for Health Statistics Rapid Surveys System, United States, October to November 2023. MMWR Morbidity and Mortality Weekly Report, 73(40), 890 to 895. https://doi.org/10.15585/mmwr.mm7340a1
• Staiman, A. (2026, February 3). Part 2: Why authors aren’t disclosing AI use and what publishers should (not) do about it. The Scholarly Kitchen. https://scholarlykitchen.sspnet.org/2026/02/03/why-authors-arent-disclosing-ai-use-and-what-publishers-should-not-do-about-it-part-2/
• Taheri, A., Tazike, M., Carrington, P., & Bigham, J. P. (2026). When disability disclosure travels: Memory, privacy, and contextual integrity in conversational AI. arXiv:2609.22720. https://arxiv.org/abs/2609.22720
• Substack. (2026, August 9). How can I detect AI on Substack? Substack Help Center. https://support.substack.com/hc/en-us/articles/50891130623508-How-can-I-detect-AI-on-Substack
• Talk Architecture. (2026, August 19). Who was the office built for? Neurodivergent spaces: The standard isn’t neutral (Season 19, Episode 9) [Podcast]. https://naziaty.buzzsprout.com/1042819/episodes/19669133-who-was-the-office-built-for-neurodivergent-spaces-the-standard-isn-t-neutral
• Tangermann, V. (2025, December 5). Woman hailed as hero for smashing Meta smart glasses on subway. Futurism. https://futurism.com/future-society/woman-hero-smashing-meta-smart-glasses-subway
• Taylor & Francis. (2025, November 19). AI policy. https://taylorandfrancis.com/our-policies/ai-policy/
• TechRadar. (2026, October 14). Some Ray-Ban Meta users are fuming after their smart glasses make secret audio recordings, but Meta says the issue is “resolved.” https://techradar.com/computing/virtual-reality-augmented-reality/some-ray-ban-meta-users-are-fuming-after-their-smart-glasses-make-secret-audio-recordings-but-meta-says-the-issue-is-resolved
• TechSpot. (2026, October 15). Meta’s smart glasses were recording conversations without the wake phrase, company admits. https://techspot.com/news/114214-meta-smart-glasses-secretly-recording-conversations-without-wake.html
• The College Fix. (2023, July 28). Professor with autism mistaken for AI bot. https://www.thecollegefix.com/professor-with-autism-mistaken-for-ai-bot
• The Open University. (2025, July 15). Generative AI for students. https://about.open.ac.uk/policies-and-reports/policies-and-statements/generative-ai-learning-teaching-and-assessment-ou-0
• The White Hatter. (2026, August 22). A draft policy for schools specific to smart glasses, wearable recording devices, and AI-enabled wearables. https://www.thewhitehatter.ca/post/a-draft-policy-for-schools-specific-to-smart-glasses-wearable-recording-devices-and-ai-enabled-wear
• TNW. (2026a, October 5). Hans Anders halts Meta Ray-Ban glasses sales in Netherlands and Belgium. https://thenextweb.com/news/hans-anders-meta-ray-ban-glasses-netherlands-belgium
• TNW. (2026b, October 5). Norway proposes temporary ban on AI glasses. https://thenextweb.com/news/norway-temporary-ban-ai-glasses
• Transportation Security Administration. (n.d.). TSA notification card: Individuals with disabilities and medical conditions. https://www.tsa.gov/sites/default/files/disability_notification_card_508.pdf
• Turnitin. (2023, June 14). Understanding the false positive rate for sentences of our AI writing detection capability. https://www.turnitin.com/blog/understanding-the-false-positive-rate-for-sentences-of-our-ai-writing-detection-capability
• U.S. Department of Health and Human Services. (n.d.). Employers and health information in the workplace. https://www.hhs.gov/hipaa/for-individuals/employers-health-information-workplace/index.html
• U.S. Department of Justice. (2010). Service animals: ADA requirements. ADA.gov. https://www.ada.gov/resources/service-animals-2010-requirements/
• U.S. Food and Drug Administration. (2026, July 13). 510(k) premarket notification K254044, Meta Platform Technologies, LLC. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K254044
• U.S. Government Accountability Office. (2002, September 9). SSA disability decision-making: Additional measures would enhance agency’s ability to determine whether racial bias exists (GAO-02-831). https://www.gao.gov/assets/gao-02-831.pdf
• Uebel, M., & Hamamra, B. (2026). Authorship after generative AI: Distributed creativity and relational responsibility. Philosophy and Technology, 39(3), 152. https://doi.org/10.1007/s13347-026-01170-w
• UK Cinema Association. (2023, January 12). Wearable technology guidance. https://www.cinemauk.org.uk/key-issues/wearable-technology/
• Università di Bologna. (n.d.). Adattamenti per esami (Servizio per gli studenti con disabilità e con DSA). https://site.unibo.it/studenti-con-disabilita-e-dsa/it/per-studenti/adattamenti-per-esami
• University of Arkansas at Fort Smith. (n.d.). Policy 601.15, Ethical and effective use of artificial intelligence (AI). https://uafs.edu/policies/academic-affairs/ethical-effective-use-of-ai-policy.php
• University of Notre Dame Department of Electrical Engineering. (2025, November). Graduate studies handbook (Section IV.B). https://ee.nd.edu/wp-content/uploads/sites/8/2025/11/EE-Graduate-Handbook-2025-AI-Included.pdf
• University of Notre Dame Department of Mathematics. (n.d.). Policy on the use of generative AI in comprehensive exams, theses, and dissertations. https://math.nd.edu/graduate/for-current-students/policy-on-the-use-of-generative-ai-in-comprehensive-exams-theses-and-dissertations/
• UploadVR. (2025, May 15). Ray-Ban Meta glasses: Meta AI detailed responses for blind and low vision users. https://www.uploadvr.com/ray-ban-meta-glasses-meta-ai-detailed-responses-blind-low-vision/
• UWV. (2025, December 22). Doelgroepregister. https://www.uwv.nl/werkgevers/werknemer-met-uitkering/werknemer-heeft-wajong-of-valt-onder-participatiewet/detail/doelgroepregister
• Vision Ireland. (2025, August 29). Meta smart Ray-Ban glasses as assistive technology. https://vi.ie/meta-smart-ray-ban-glasses-as-assistive-technology/
• VR.org. (2026, August 10). UK venues ban Meta smart glasses: Wetherspoons, Soho House. https://vr.org/articles/uk-venues-ban-meta-smart-glasses-wetherspoons-soho-house-2026
• W3C. (2026, March 28). Accessibility of machine learning and generative AI (Editor’s Draft). Accessible Platform Architectures Working Group. https://w3c.github.io/ai-accessibility/
• Wang, Y., Kameswaran, V., & Kacorri, H. (2025). A taxonomy of algorithmic harms for disability. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 8(3). https://doi.org/10.1609/aies.v8i3.36745
• Weedston, L. (2025, December 4). Woman breaks man’s Meta glasses on subway. Daily Dot. https://www.dailydot.com/culture/woman-broke-meta-glasses-subway/
• White House. (2025, July). Winning the race: America’s AI action plan. https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf
• Younker, S. (2025, December 15). This major cruise line just banned Meta Ray-Ban and other smart glasses. Tom’s Guide. https://www.tomsguide.com/computing/smart-glasses/this-major-cruise-line-just-banned-meta-ray-ban-and-other-smart-glasses-is-this-category-already-doomed
• Zhao, Y., Yao, Y., Fu, J., & Zhou, N. (2023). “If sighted people know, I should be able to know”: Privacy perceptions of bystanders with visual impairments around camera-based technology. In Proceedings of the 32nd USENIX Security Symposium. https://www.usenix.org/conference/usenixsecurity23/presentation/zhao-yuhang
• Zhuang, K. V., & Goggin, G. (2026). Governing generative AI for disability and inclusion. Information, Communication and Society, 29(6), 1954 to 1970. https://doi.org/10.1080/1369118X.2025.2570735

Cases

• Anderson v. TikTok, Inc., No. 22-3061 (3d Cir. Aug. 27, 2024). https://www2.ca3.uscourts.gov/opinarch/223061p.pdf
• Cedeno v. Walt Disney Parks and Resorts U.S., Inc., No. 6:25-cv-02046-GAP-DCI (M.D. Fla.). Docket entries 28 (notice of voluntary dismissal, October 16, 2026) and 29 (order dismissing case, October 20, 2026), as listed in the CourtListener docket feed. https://www.courtlistener.com/docket/71732203/bruno-cedeno-v-walt-disney-parks-and-resorts-us-inc/
• Cleveland v. Policy Management Systems Corp., 526 U.S. 795 (1999). https://supreme.justia.com/cases/federal/us/526/795/
• Moody v. NetChoice, LLC, 603 U.S. 707 (2024). https://openjurist.org/603/us/707
• Texas v. Kennedy (filed as Texas v. Becerra), No. 5:24-cv-00225 (N.D. Tex.). (2026a). Corrected joint motion for entry of final judgment, Doc. 124 (September 1, 2026). https://dredf.org/wp-content/uploads/2026/09/124.-Corrected-Jt-Mot-for-Entry-Final-J-Access-Pass.pdf
• Texas v. Kennedy (filed as Texas v. Becerra), No. 5:24-cv-00225 (N.D. Tex.). (2026b). Order and final judgment, Doc. 125 (September 23, 2026). https://storage.courtlistener.com/recap/gov.uscourts.txnd.395028/gov.uscourts.txnd.395028.125.0.pdf
• XXXX v. HR Rail SA, Case C-485/20, ECLI:EU:C:2022:85 (CJEU, February 10, 2022). https://curia.europa.eu/juris/liste.jsf?num=C-485/20

Statutes, regulations, and instruments

United States

• Americans with Disabilities Act, 42 U.S.C. § 12101 (findings and purpose). https://www.law.cornell.edu/uscode/text/42/12101
• Americans with Disabilities Act, 42 U.S.C. § 12102 (definition of disability). https://www.law.cornell.edu/uscode/text/42/12102
• Americans with Disabilities Act, 42 U.S.C. § 12111 (definitions, Title I). https://www.law.cornell.edu/uscode/text/42/12111
• Americans with Disabilities Act, 42 U.S.C. § 12112. https://www.law.cornell.edu/uscode/text/42/12112
• Americans with Disabilities Act, 42 U.S.C. § 12117 (enforcement, Title I). https://www.law.cornell.edu/uscode/text/42/12117
• Americans with Disabilities Act, 42 U.S.C. § 12133 (enforcement, Title II). https://www.law.cornell.edu/uscode/text/42/12133
• Americans with Disabilities Act, 42 U.S.C. § 12188 (enforcement, Title III). https://www.law.cornell.edu/uscode/text/42/12188
• Americans with Disabilities Act, 42 U.S.C. § 12201(h) (regarded-as coverage and accommodation). https://www.law.cornell.edu/uscode/text/42/12201
• Americans with Disabilities Act, 42 U.S.C. § 12203 (retaliation and coercion). https://www.law.cornell.edu/uscode/text/42/12203
• Americans with Disabilities Act, 42 U.S.C. § 12205 (attorney’s fees). https://www.law.cornell.edu/uscode/text/42/12205
• Civil Rights Act of 1964, Title VII, 42 U.S.C. § 2000e-5 (enforcement provisions, adopted for Title I by § 12117). https://www.law.cornell.edu/uscode/text/42/2000e-5
• Civil Rights Act of 1991, 42 U.S.C. § 1981a (damages in cases of intentional discrimination). https://www.law.cornell.edu/uscode/text/42/1981a
• Communications Decency Act, 47 U.S.C. § 230. https://www.law.cornell.edu/uscode/text/47/230
• Family Educational Rights and Privacy Act, 20 U.S.C. § 1232g. https://www.law.cornell.edu/uscode/text/20/1232g
• Rehabilitation Act of 1973, Section 504, 29 U.S.C. § 794. https://www.law.cornell.edu/uscode/text/29/794
• Rehabilitation Act of 1973, 29 U.S.C. § 794a (remedies and attorney fees). https://www.law.cornell.edu/uscode/text/29/794a
• Secure 2.0 Act of 2022, Section 124, Division T of Public Law 117-328. https://www.govinfo.gov/content/pkg/PLAW-117publ328/html/PLAW-117publ328.htm
• 16 C.F.R. Part 318 (Health Breach Notification Rule). https://www.ecfr.gov/current/title-16/chapter-I/subchapter-C/part-318
• 20 C.F.R. §§ 404.1520, 404.1571, 404.1572, 404.1574, 404.1576. https://www.ecfr.gov/current/title-20/chapter-III/part-404/subpart-P
• 23 C.F.R. Part 1235 (Uniform System for Parking for Persons with Disabilities). https://www.ecfr.gov/current/title-23/chapter-II/part-1235
• 26 C.F.R. § 1.529A-2 (Qualified ABLE program). https://www.law.cornell.edu/cfr/text/26/1.529A-2
• 28 C.F.R. §§ 35.104, 35.160 (auxiliary aids and services; effective communication). https://www.law.cornell.edu/cfr/text/28/35.104 ; https://www.law.cornell.edu/cfr/text/28/35.160
• 28 C.F.R. §§ 35.130, 36.302. https://www.ecfr.gov/current/title-28/chapter-I/part-35/subpart-B/section-35.130 ; https://www.ecfr.gov/current/title-28/chapter-I/part-36/subpart-C/section-36.302
• 28 C.F.R. § 35.136 (service animals, Title II). https://www.law.cornell.edu/cfr/text/28/35.136
• 28 C.F.R. §§ 35.139, 36.208 (direct threat). https://www.law.cornell.edu/cfr/text/28/35.139 ; https://www.law.cornell.edu/cfr/text/28/36.208
• 28 C.F.R. § 35.170 (complaints, Title II). https://www.law.cornell.edu/cfr/text/28/35.170
• 28 C.F.R. § 36.303 (auxiliary aids and services, Title III). https://www.law.cornell.edu/cfr/text/28/36.303
• 29 C.F.R. § 1630.2 (definitions). https://www.law.cornell.edu/cfr/text/29/1630.2
• 29 C.F.R. § 1630.14. https://www.ecfr.gov/current/title-29/subtitle-B/chapter-XIV/part-1630/section-1630.14
• 34 C.F.R. Part 99 (Family Educational Rights and Privacy). https://www.ecfr.gov/current/title-34/subtitle-A/part-99
• 34 C.F.R. §§ 361.5, 361.42. https://www.law.cornell.edu/cfr/text/34/361.5 ; https://www.law.cornell.edu/cfr/text/34/361.42
• 38 C.F.R. §§ 4.1, 4.16. https://www.law.cornell.edu/cfr/text/38/4.1 ; https://www.law.cornell.edu/cfr/text/38/4.16
• 45 C.F.R. §§ 164.502, 164.514, 164.524. https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-C/part-164/subpart-E/section-164.502 ; https://www.law.cornell.edu/cfr/text/45/164.514 ; https://www.law.cornell.edu/cfr/text/45/164.524
• Wash. Rev. Code ch. 19.373 (My Health My Data Act). https://app.leg.wa.gov/RCW/default.aspx?cite=19.373

European Union and member states

• Council Directive 2000/78/EC of 27 November 2000 establishing a general framework for equal treatment in employment and occupation. https://eur-lex.europa.eu/eli/dir/2000/78/oj
• Regulation (EU) 2016/679 (General Data Protection Regulation), Articles 5, 9, 22, 77, and Recital 35. https://eur-lex.europa.eu/eli/reg/2016/679/oj
• Directive (EU) 2019/882 (European Accessibility Act). https://eur-lex.europa.eu/eli/dir/2019/882/oj
• Regulation (EU) 2022/2065 (Digital Services Act), Articles 16, 17, 19, 20, and 21. https://eur-lex.europa.eu/eli/reg/2022/2065/oj
• Regulation (EU) 2024/1689 (Artificial Intelligence Act), as amended by Regulation (EU) 2026/1744. https://eur-lex.europa.eu/eli/reg/2024/1689/oj ; https://eur-lex.europa.eu/eli/reg/2026/1744/oj
• Directive (EU) 2024/2841 (European Disability Card and European Parking Card). https://eur-lex.europa.eu/eli/dir/2024/2841/oj
• Sozialgesetzbuch Neuntes Buch (SGB IX), §§ 2, 152. https://www.gesetze-im-internet.de/sgb_9_2018/__2.html ; https://www.gesetze-im-internet.de/sgb_9_2018/__152.html
• Real Decreto Legislativo 1/2013, de 29 de noviembre, Ley General de derechos de las personas con discapacidad y de su inclusión social, as amended by Ley 4/2026 (see BOE, n.d.; BOE, 2026).
• Ley 39/2015, de 1 de octubre, del Procedimiento Administrativo Común de las Administraciones Públicas, Artículo 16. https://www.boe.es/buscar/act.php?id=BOE-A-2015-10565

Frequently Asked Questions

What is the AI Disability Standard?

The AI Disability Standard is the system the paper calls for: a notice a person can send today, with a validation behind it that no one issues yet. The notice invokes protections that already exist, and the validation lets a recipient see the person’s standing without opening a medical file.

What are the three punishments the paper names?

The paper names refusal, algorithmic punishment, and the label. Refusal is a rule that bars the tool, a venue that bars the device, or a platform that bars the use. Algorithmic punishment is a detector, a ranking system, or a classifier acting on how the work reads. The label is a requirement to mark the work as AI-assisted.

Does the notice create a legal right?

No. The notice and the validation create no right. Every protection the notice names exists before the letter is written, and the letter’s work is to make the recipient knowing. Where the law reaches the recipient, force means a duty to consider and respond, never a duty to grant the means named.

What does the notice contain?

A notice carries nine elements, in order: the addressee, standing and means, the policy or action at issue, the protections it conflicts with, the request, the boundary, confidential handling, the record, and a reservation of rights. The person states the barrier, what the AI does about it, and what that allows, and states no diagnosis.

Where does the notice have force today?

The notice has force against covered employers, funded programs, and public entities, where the sender is a person the statute protects and the need is communicated. In Spain it reaches any recipient of the right to use support products without forced deactivation. Against platforms and publishers no federal or Union-wide law binds the recipient, so the notice builds a record.

Which disability systems does the validation borrow from?

The validation borrows from six disability systems already in use. They are the Social Security Administration’s functional definition, SSDI receipt as the proposed automatic qualifier, the ABLE account’s certification, the parking placard’s entitlement-only face, the TSA Notification Card’s discreet presentation, and HIPAA’s minimum necessary principle for the notice itself.

Why are AI glasses the stress test?

One device carries every strand at once. The camera that reads a menu for a legally blind wearer is the camera bystanders fear, so a single pair of glasses raises the accessibility use, the privacy outcry, denial of access, hostility toward wearers, and regulators acting on both sides.

What does the paper ask of the disability community?

It asks 28 questions and settles none of them, from where the eligibility line should sit to who issues a validation and where records live. It carries the full case against its own proposal and asks the community to design the validation, or reject it and say what should stand in its place.

Disclaimer

The author is not a lawyer, and this paper does not provide legal advice. This is thought research and governance analysis based on public sources, cited materials, and human-AI review. It is written for people who use AI as a disability tool, for those who advocate for them, and for the accommodation offices, platforms, publishers, and policymakers whose rules reach them. Readers should not rely on this paper as a legal opinion, compliance determination, or substitute for qualified counsel. A person considering a notice, and an organization receiving one, should consult an attorney in their own jurisdiction before acting.

The Other AI: Audio Briefings on Augmented Intelligence and AI Governance

Spotify | Apple Podcasts | Amazon Music | YouTube Playlist

#AIassisted using the HAIA Ecosystem | CC BY-NC-SA 4.0

Free for personal, educational, and noncommercial research use with attribution. Commercial exploitation, paid productization, and enterprise commercialization require separate permission and licensing.

Suggested citation: Puglisi, B. C. (2026). A Call for an AI Disability Standard. basilpuglisi.com.