Should accessibility law protect a right to use AI as an assistive tool?

The policy choice is about access, not loyalty to a product

The strongest version of the proposal is not “everyone has a right to ChatGPT,” or any other named service. It is: a disabled person should be able to use, or obtain, an effective assistive method when it is necessary for equal access and does not defeat a legitimate purpose of the activity. AI can be one such method, alongside screen readers, speech recognition, captioning, alternative formats, human interpreters, scribes, calculators, and adapted equipment.

That framing matters because AI is not a single thing. A local text predictor that helps a person with dysgraphia draft an email has different risks from a cloud service that stores confidential case records, or an autonomous agent that sends messages and changes files. A rule tied to an outcome can ask what functional barrier the person faces, what result the activity is supposed to measure, and whether an assistive method safely removes the barrier. A statute tied to the word “AI” could accidentally protect a product category while excluding the next useful technology.

The proposal still has practical force. It tells an institution that “our policy bans AI” is not the end of the inquiry when the tool is requested as disability support. The institution should assess the request, identify the core purpose of the task, consider a privacy-preserving or supervised configuration, and offer an effective alternative if the exact tool cannot be allowed. It also tells technology providers and public procurers to build systems that work with assistive tools, rather than treating a disabled person's workflow as an unsupported exception.

What current law already provides

The following comparison concerns only the United States and the European Union. It is a high-level policy map, not legal advice. National, state, local, sectoral, and contractual rules can add obligations or remedies.

In the United States, the employment rule is not a right to the worker's favourite technology. The United States EEOC says the governing test is effectiveness, and an employer need not lower quality or quantity standards or provide an accommodation that causes undue hardship. United States EEOC ADA Questions and Answers

In the United States, public entities must give people with disabilities an equal opportunity to benefit from state and local government programs and make necessary reasonable modifications unless they can demonstrate fundamental alteration. The United States Department of Justice also states that public entities are not required to take actions that create undue financial and administrative burdens, although the analysis does not end simply because the first requested solution is unavailable. United States ADA Title II guide

In the European Union, the European Accessibility Act is already an example of a law that regulates accessibility outcomes without naming AI. It covers particular products and consumer services, including consumer general-purpose computer hardware and operating systems, e-commerce services, and other listed categories. Its Annex I includes functional performance criteria for functions not otherwise addressed, while Article 14 recognises fundamental alteration and disproportionate-burden limits that operators must assess and document. European Union Accessibility Act, scope and Article 4 · European Union Accessibility Act, Article 14

The strongest case for treating AI as assistive technology

AI can reduce disability-related friction where ordinary software does not. It can turn dictated speech into structured text, simplify a dense document while preserving the original alongside it, describe a complex visual in a chosen level of detail, support drafting, summarize a long meeting transcript, or convert a person's preferred communication method into a format another person can use. For a person who must otherwise spend substantially more time or use a human intermediary, the gain can be access to the activity itself rather than mere convenience.

The strongest case is therefore one of functional equivalence. A person should have a fair chance to learn, work, take part in civic processes, or obtain a service. If an AI feature supplies reading, writing, communication, organization, or navigation support that others obtain through ordinary interfaces, a blanket ban may perpetuate the barrier that accessibility law is designed to address.

AI can also make individualized support more available. A human reader, interpreter, coach, or scribe can be indispensable, but may be expensive, unavailable at the needed time, or intrusive. An assistive AI option can improve privacy and independence when it is voluntary, understandable, and used within an appropriate data-governance framework. This is an argument for adding a capable option, not for replacing human support that a person needs or prefers.

Example

Hypothetical United States employment case: A policy analyst with a disability that substantially limits written expression asks to use an employer-approved AI drafting assistant. The role requires accurate policy judgment, source checking, and final accountability, but not unaided first-draft composition. The employer configures the tool so confidential records stay in an approved environment, requires the analyst to verify citations and sign the final product, and audits use only as necessary for security. The AI assists with a disability-related barrier while the analyst still performs the role's essential judgment.

The takeaway is not that the worker has a right to any consumer chatbot. In the United States, the relevant question is whether the approved configuration or another accommodation effectively enables the worker to perform essential functions without undue hardship. The EEOC's guidance gives examples including modified equipment, readers, interpreters, and modified training or examinations, and says the choice is fact-specific. United States EEOC ADA Questions and Answers

Why a named AI right would be too broad

An unlimited technology-specific right would collide with real and sometimes important interests. These limits should be assessed with evidence and alternatives, not used as an excuse for a reflexive ban.

The table points to an important distinction. A privacy or assessment-validity concern may justify declining one tool configuration. It does not automatically justify denying the person access to the underlying program, work, service, or exam. The institution should ask whether a local model, an enterprise account, a locked-down feature, extra time, a human reader, a scribe, or a redesigned accessible interface can achieve comparable access.

How the rule would work in four settings

Education in the United States

For United States K-12 special education, the Individuals with Disabilities Education Act already offers a better model than a product mandate. The IEP team must consider assistive technology for every child with an IEP. If the team determines it is required for a free appropriate public education, the local educational agency must provide and maintain the device and necessary service. The guidance leaves the type of technology to an individualized determination of need. United States Department of Education IDEA assistive-technology guidance

Hypothetical United States education case: A student with a reading disability asks to use an AI reading and vocabulary assistant for history homework. The school tests an approved version, limits data sharing, gives the student access to the original text, and records how the tool supports the IEP goal. If the tool proves effective, it may become a supported assistive technology service. If the same student takes a test of independent close reading, the school can use a different accessible format and accommodations that preserve the test's purpose.

For United States schools and colleges using cloud AI with education-record information, access does not erase privacy obligations. United States Department of Education guidance says an online service receiving personally identifiable information under FERPA's school-official exception must perform a service the school would otherwise use staff for, remain under the school's direct control for use and maintenance of the data, and avoid unauthorized redisclosure or use. United States Department of Education FERPA guidance

Employment in the United States

For United States employment, an employee who needs AI support should make a disability-related accommodation request through the employer's process. The employee may suggest a particular tool, but the employer and employee should identify an effective accommodation through an interactive process. The United States EEOC says an employer may need to modify equipment or provide other adjustments, but is not required to lower performance standards, remove essential functions, or accept undue hardship. United States EEOC accommodation guidance

Hypothetical United States employment case: A customer-service employee with a neurological disability asks to use real-time AI speech-to-text and text-rephrasing support. The employer can assess whether the customer-consent, data-security, and accuracy requirements can be met. If direct use of a public service would expose customer information, the employer should examine an approved private deployment, a different accessibility tool, or human support. A bare “we ban AI” policy may be inadequate if it does not consider effective alternatives.

The same rule protects legitimate boundaries. A United States employer may require the employee to comply with quality, confidentiality, conflict-of-interest, and final-review standards that apply to the job. Accessibility should remove a disability-related barrier, not transfer the role's ultimate accountability to software.

Examinations in the United States

For United States admissions, licensing, certification, and other covered high-stakes tests, the central question is the skill the exam is designed to measure. United States Department of Justice guidance says testing entities must ensure scores reflect aptitude, achievement, or the intended skill rather than an unrelated impairment. It gives the familiar example that a calculator can be suitable on an algebra test when basic computation is secondary, but not when computation itself is the target. United States ADA testing-accommodations guidance

Hypothetical United States examination case: A professional licensing exam tests knowledge of safety standards, not spelling or handwriting. A candidate with a disability requests a restricted AI speech-to-text tool that cannot retrieve outside information and produces only a transcript of dictated answers. That may be easier to justify than an unrestricted generative AI assistant that supplies substantive answers. The testing entity should assess the requested configuration and any equally effective alternative, not apply a label-based “AI is cheating” rule.

Where the assessment is designed to measure the very work the AI would perform, limiting the AI can be legitimate. The governing principle is not whether the technology is impressive or popular. It is whether the accommodation preserves a valid measurement while giving the candidate an equal opportunity to demonstrate the intended skill.

Public services in the United States and European Union

For United States state and local government services, ADA Title II covers public benefits, courts, public education, transportation, voting, and other programs. A public entity must provide equal opportunity and make necessary reasonable modifications unless it can demonstrate fundamental alteration. Since 24 April 2026, larger United States public entities generally have a web and mobile-app accessibility compliance date under the Department of Justice Title II rule, with a later 2027 date for smaller entities and special districts. That rule is about accessible public web content and apps, not a personal entitlement to use generative AI. United States ADA Title II web and mobile regulation

Hypothetical United States public-service case: A county benefits portal is compatible with screen readers, but a resident with a cognitive disability needs an approved plain-language AI assistant to understand a complex renewal notice and complete the application. The county should consider a privacy-protective assistant, clear human support, or another effective method. If it cannot allow the resident's preferred public AI account because it would disclose benefit data, it should explain the risk and provide an effective alternative rather than end the inquiry.

For European Union covered consumer services and products, the European Accessibility Act requires specified accessibility outcomes, including a design that supports foreseeable use by persons with disabilities. It is not a general individual accommodation statute and is implemented by Member States, but it demonstrates why functional performance criteria can outlast individual interface fashions. European Union public-sector websites and mobile applications are also addressed by the separate Web Accessibility Directive, which requires accessible public-sector web and mobile content subject to a disproportionate-burden rule and accessible alternatives where that rule is invoked. European Union Accessibility Act · European Union Web Accessibility Directive

A better legislative design

A new law or an amendment could create a right to an effective assistive method, with AI expressly included as a possible method but not guaranteed as a named product. The following design choices would make the rule clearer and more durable.

1. Define the right by outcome. Require equal access, effective communication, or a reasonable accommodation that removes a disability-related barrier. State that AI may be an assistive technology when effective for the person and task.

2. Require an individualized process. The user should be able to request a tool or explain the barrier. The institution should reply promptly, assess the particular configuration, and give written reasons if it declines the request. A categorical ban should trigger review, not end the process.

3. Preserve effective alternatives. The person should not automatically control brand, deployment, or model. The provider may choose an equally effective option, but must be able to show that it actually works for the individual and the activity.

4. Protect validity without stereotype. In education, licensing, and employment, define the essential skill or assessment construct before deciding whether AI can assist. Use limited modes, separate assessment formats, or human supervision where needed. Do not assume that every AI use creates unfair advantage.

5. Set privacy and security floors. Require data minimisation, transparency about processing, usable consent where appropriate, accessible privacy notices, security review, retention limits, and no covert use of disability data for unrelated profiling or model training.

6. Control consequential actions. AI may draft, translate, summarize, or navigate. For legal, medical, financial, employment, and public-benefit decisions, require the accountable person or system to verify the result and preserve an appeal or correction path.

7. Make procurement and interoperability part of access. Public bodies and large institutions should procure systems compatible with common assistive technologies and provide interfaces, export formats, and API policies that do not arbitrarily block a disabled person's approved tool.

8. Provide review and remedy. Require accessible request channels, timely decisions, a way to challenge a denial, and collection of privacy-preserving data about failures. Without a process and remedy, a nominal right can be too expensive to use.

This design does not require an institution to build an AI assistant for every person. It creates a disciplined process: identify the barrier, assess the requested method, explain any concrete limit, and deliver an effective alternative where required. It also prevents an AI-specific rule from becoming obsolete when the next assistive interface arrives.

Common policy mistakes

• Treating AI as one risk category. Text prediction, local transcription, cloud summarization, and an autonomous agent have different data, accuracy, and action risks. Regulation should distinguish them.

• Treating an access request as a request for a consumer subscription. The legal and practical issue is often a safe configuration, not reimbursement for a particular product.

• Using privacy as a blanket refusal. Privacy can be a decisive constraint, especially for student, health, employment, or benefit data. It should prompt an alternative design, not erase the access obligation.

• Ignoring the assessment's purpose. An unrestricted generative tool can invalidate an exam or credential if it performs the target skill. The answer may be a constrained tool, a different format, or a different measure, not necessarily a complete ban.

• Assuming automation solves accessibility. An AI interface that hallucinates, cannot be navigated with a screen reader, or exposes sensitive data can create a new barrier. Human support and accessible non-AI routes remain necessary.

• Making the individual prove every technical detail. The requester can describe the barrier and the desired support. The institution normally has more control over its systems, procurement, security review, and alternatives.

Decision checklist for an institution

• Name the jurisdiction and the applicable accessibility, education, employment, testing, privacy, and sectoral rules.
• Identify the functional barrier and the activity the person needs to perform.
• Identify what the activity is meant to measure or achieve, including any essential job function or assessment construct.
• Describe the requested AI configuration, including provider, data flow, retention, tool access, and human review.
• Test whether the method is effective for the person, rather than relying on a general AI policy.
• Assess security, privacy, licensing, safety, integrity, and cost with evidence.
• Consider an equally effective alternative and explain why it is genuinely comparable.
• Record reasons for any fundamental-alteration, undue-burden, or disproportionate-burden conclusion where the relevant law permits one.
• Provide an accessible appeal, review, or accommodation process.
• Revisit the decision when the person's needs, the tool, the service, or the law changes.

Limits of this answer

This is policy analysis, not personalized legal advice. In the United States, the ADA's coverage, applicable title, employer size, the nature of a testing entity, federal-funding status, state law, collective agreements, and particular facts can all matter. In the European Union, the European Accessibility Act's scope, national transposition and enforcement, sector-specific law, and the Member State involved can all matter. A person considering a complaint or an institution considering a refusal should obtain advice from a qualified disability-rights, education, employment, privacy, or regulatory professional in the relevant jurisdiction.