
AI access · Disability rights
AI access is a disability rights issue
Human contribution, AI assistance, and the duty to accommodate.
A little help can change what is possible. An idea finds words. A story reaches a reader. Someone who has struggled to be heard gets to speak in their own way.
AI can be part of that help. The person still brings the experience, the purpose and the choices.
Our rules need to make room for this. We can protect privacy, recognize people’s work and make participation easier. We don’t have to choose between care and possibility.
This is the question behind our latest ABILITYscaled PSA: how do we turn a useful tool into a real opportunity, rather than another hurdle? The barriers deserve attention. So do the ways through.
9 pages · 279 KB
Participation is the point
Everyone has something to say, create or contribute. For people with disabilities, AI assistance can help turn those possibilities into everyday realities: being understood, sharing knowledge, pursuing meaningful work and having a say in decisions that shape their lives. The person’s experience, ideas and choices remain at the heart of that contribution.
Participation should not come at the cost of privacy or require people to prove they can manage without assistance. Rules should protect people without closing doors. When they create a barrier, people deserve a clear explanation, someone who will listen and a practical way forward. The stories that follow show what assistance can make possible. The legal principles explain why institutions must take disability-related barriers seriously.
What assistance makes possible
Edis builds a tool for blind makers
Edis, a blind accessibility consultant, reports using Claude to build AccessSlice for accessible 3D printing despite not being a professional programmer. His knowledge guided the work; AI helped supply the programming. Successful prints on four printer models are reported by the creator. [12]
Luke Stoffel keeps complex creative work moving
Writer Luke Stoffel, who works with dyslexia and ADHD, describes using AI to organize his archive and sustain complex creative work. His book Boy Refracted is independently documented; AI’s role is his own account. [13, 14]
Artists conserve energy for creative work
A study of five disabled visual artists found that AI-assisted administration could conserve energy for creative work, alongside concerns about cost, interfaces and personal voice. These accounts show how assistance may sustain participation, not how widely people benefit. [15]
What participation can look like
A tenant needs a repair rather than a writing test
A tenant uses AI to put dates in order and explain a repair request, then checks that it is accurate. The housing provider needs reliable information, not proof that the tenant wrote every word without help. Rejecting the request because AI helped with the wording leaves the repair unresolved. What matters is whether the tenant’s living conditions improve.
An employee needs support that makes work sustainable
An employee uses an approved AI tool to help turn their knowledge into a report. They check the result and take responsibility for it. The report must still be accurate, and confidential information must stay protected. The benefit is being able to keep working and earning an income with less exhausting effort. If checking the report takes more effort than the tool saves, that benefit may disappear.
A student needs a fair way to show understanding
AI can help a student put their ideas in order. In a history assignment, talking through their answer and checking their sources may show what they understand. If the assignment tests writing without help, using AI changes what the work shows. The rules should match what the student is being asked to learn, while giving them an accessible way to show it.
A patient needs their own account to be understood
A patient uses AI to organize symptoms and questions or prepare a plain-language draft for discussion with a clinician. Important information needs checking because AI can omit details or introduce errors. The aim is less confusion and fuller participation in decisions.
Getting help does not erase your contribution
The person’s purpose, experience and judgment remain central when AI helps with a task. Requiring someone to show they can work without help before allowing assistance creates another barrier. People still need accessible ways to guide and check AI’s work because it can introduce mistakes or change what they mean. For publishing, there are two separate questions: whether assistance enables participation, and whether copyright exists in every AI-generated part. Taking responsibility for a publication does not settle the copyright question. [4]
Disability rights must shape the rules
Where the legal protections come from
These protections come from different sources. The UN disability rights convention sets out Canada’s international obligations. Canadian human-rights laws prohibit discrimination and require accommodation within their fields of application. The Accessible Canada Act adds requirements to identify, remove and prevent barriers within federal jurisdiction. Privacy laws protect personal information. Which law applies and where a person can challenge a decision depend on the institution and activity involved. [3, 5, 6, 9, 10, 11, 16]
When a rule discriminates
A rule can be the same for everyone and still discriminate against someone because of disability. To establish discrimination under the applicable human-rights law, a person must show a disability, a disadvantage in an activity the law protects, and a connection between the two. The institution must then establish a lawful justification. When it relies on a necessary job requirement or service rule, that includes showing it cannot accommodate the person without undue hardship. Under the federal Canadian Human Rights Act, health, safety and cost are the factors considered. Good intentions do not remove that responsibility. [5, 17]
Support must work in practice
The question is whether people can meaningfully access the service, not whether support is treated as a luxury. In Moore, the Supreme Court explained that disability supports can be the means of access to education. Grismer rejected an inflexible disability-related driving standard applied without individual assessment. Eldridge required sign-language interpretation where necessary for effective communication in publicly funded medical care. These cases show why institutions must examine actual barriers and accommodation. They did not decide AI access or guarantee a particular tool. Charter duties apply where the decision involves government action or implementation of a government program. [17, 19, 20]
An alternative must provide effective support in practice, not merely exist on paper. The UN disability rights convention connects equality and participation with support for people’s choices, privacy, education and health. Canadian law determines the duties and remedies that apply to a particular decision. [3, 11]
AI accessibility is already a policy priority
Employment and Social Development Canada is exploring AI in Phase 2 of its digital-accessibility regulatory work. Its backgrounder recognizes that AI can improve access but also create barriers, and emphasizes involving people with disabilities in its development. Existing rights obligations still apply. [21]
Do publishing rules recognize the person’s contribution?
AI helped us develop this publication. Library and Archives Canada’s (LAC) ISBN guidance provides a concrete example of how institutions assess assistance. An ISBN identifies an edition and format for booksellers and libraries; it does not settle copyright, attribution or factual reliability. [1, 2]
The guidance does not explain where the line is drawn
Under “Use of artificial intelligence,” LAC states: “By registering with ISBN Canada, you declare that your publication was written by a human author. You may use AI tools only for limited support tasks, such as brainstorming ideas, checking grammar, or formatting your book. ISBN Canada limits the quantity of ISBNs for works whose content is primarily generated by AI (text, illustration, translation). LAC is currently reviewing its policy regarding the use of AI content creation and our guidance is subject to change.” [1]
The guidance does not define “limited support tasks” or “primarily generated by AI,” or explain how those categories are assessed. It combines a human-author declaration and a restriction to limited AI support with quantity limits for primarily AI-generated content. It does not explain how these provisions work together. That uncertainty matters when assistance enables a person’s contribution.
ABILITYscaled has completed publisher registration. We have not confirmed a title-specific ISBN for this publication. This is a concern about the public rules, not a report of an ISBN refusal.
A person contributes more than the words they type
A person may choose a publication’s purpose, develop its argument, check evidence and direct revisions while using AI to help produce the wording. A distinction based on whether content is primarily generated by AI may not capture those contributions. Institutions need to distinguish concerns about credit, quality and copyright from assistance that makes earning a living, public participation or creative work possible. ISBN rules do not govern those other activities. The wider question is whether each rule addresses a real risk without unnecessarily excluding the person.
Protect people without closing doors
We recommend safeguards that fit the task and the risk: protect confidential information, check important claims, explain relevant AI assistance and assess the knowledge or skill the task actually requires. Explaining how AI helped should not unnecessarily expose someone’s disability, private drafts or full history of assistance. Helping a person express a decision is not permission to decide or act for them. The UN disability rights convention protects privacy and supports people in exercising their own choices. [3]
Children must remain part of the conversation
An answer that sounds polished may not reflect what a child wants to say. Children need accessible ways to make choices, disagree and change their answers. The UN disability rights convention calls for assistance suited to their disability and age so they can express their own views. Protection and guidance must respect their growing ability to make decisions and keep their voice central. [3, 7, 8]
Canada’s Children’s Privacy Code explains how the Personal Information Protection and Electronic Documents Act (PIPEDA) protects children’s personal information. It calls for clear, accessible information, support suited to their growing capacity, and practical ways to correct personal information. [18]
A way to put things right
When a decision creates a barrier, we recommend clear reasons, an accessible review process and timely correction. An opportunity may be lost while someone waits; feedback is not enough if it cannot lead to reconsideration. The available legal remedy depends on the law and decision involved, so these recommendations do not establish a universal right to review. Each participant remains responsible for the decisions they control. In particular, a person’s duty to check AI’s work does not remove the provider’s obligations under applicable privacy law. [22]
The May 2026 joint Canadian privacy investigation of OpenAI illustrates those responsibilities. It examined consent, accuracy of personal information, access, correction, deletion and accountability under federal and provincial private-sector privacy laws. The inquiry focused on GPT-3.5 and GPT-4 and considered subsequent protective measures. The investigation shows why safeguards cannot rest on users alone. [22]
Protect privacy while learning what works
Learning whether assistance helps should not require unnecessary personal disclosure. Keep optional research separate from decisions about who qualifies for support, and examine reasons for restrictions, correction rates and waiting times before routinely seeking diagnoses, private drafts or complete chat histories. For federal institutions, section 4 of the Privacy Act requires personal information collected to relate directly to a program or activity; other sectors have different rules. [9] Accessible, voluntary accounts may reveal experiences missing from application records. They help explain what happens to people, but cannot establish how many in the wider population benefit.
Keep the choice to use other support
People should be free to use AI, guide how it helps or decline it. Human support, accessible documents and ways to communicate without AI must remain available. Accessibility Standards Canada’s voluntary standard also emphasizes people’s control and access to alternatives. [10]
Are people’s lives getting better?
These stories show what AI can make possible. We still need to understand how widely the benefits are shared and whether they last. The measure is what changes in people’s lives.
Follow what matters to the person
Start with what matters to the person: secure housing or income, a project or opportunity they can sustain, less time spent on paperwork, creative fulfilment or a greater say in decisions. Look at their situation before assistance, the help they receive and what happens afterwards. Follow long enough to see whether improvements last. Human support and institutional decisions may also shape the result. Record mistakes, the effort needed to check work, costs, privacy losses and exclusion as well as benefits.
The ambition is a better life
At ABILITYscaled, we expect AI to help a great many people with disabilities live better lives, with improvements that can be measured. We believe its importance could surpass earlier advances, apart from the UN Convention on the Rights of Persons with Disabilities (UNCRPD). That is our conviction about the future, not a proven comparison with the past. AI builds on earlier supports rather than replacing them. The distinction is that the UNCRPD establishes obligations to respect rights, while AI may help people pursue ambitions that were previously out of reach.
Keep participation possible
Protect people from genuine risks without closing the doors that make their contribution possible. Institutions should explain the risk a rule addresses, examine the disability-related barriers it creates, consider accommodation and offer an accessible way to review decisions. We also need evidence that assistance helps people pursue opportunities and live better lives. The aim is meaningful participation, with safeguards and accountability that make it possible.
Sources for the stories and analysis
Sources were reviewed between 21 and 26 October 2026. The quoted LAC guidance is from its About ISBNs page, dated 9 October 2026. Numbered references identify legal authorities, research, first-person accounts and regulatory or policy guidance, with linked source titles. Illustrative situations and recommendations are ABILITYscaled’s analysis.
1. Library and Archives Canada. About ISBNs. “Use of artificial intelligence” and “About copyright.” Page date modified: 9 October 2026. The AI wording quoted in this publication is from that page. LAC states that its AI content-creation policy is under review and its guidance is subject to change. An ISBN does not provide copyright protection.
2. International ISBN Agency. What is an ISBN. Identifier purpose and publication formats.
3. United Nations. Convention on the Rights of Persons with Disabilities. Articles 4(3), 5, 7(3), 9, 12, 21, 22, 24, 25, 27, 30(2) and 31. Article 5 addresses equality and reasonable accommodation; Article 12 addresses support in exercising legal capacity; Article 22 protects privacy; Articles 24 and 25 concern education and health. Article 4(3) requires close consultation and active involvement through representative organizations. Article 31 concerns safeguarded information used to assess implementation and address barriers; it is distinct from CRC Article 31 on cultural and artistic participation.
4. Supreme Court of Canada. CCH Canadian Ltd v Law Society of Upper Canada. 2004 SCC 13. Originality requires non-trivial skill and judgment. The case did not decide generative-AI authorship; the distinction between publication responsibility and copyright in a generated element is this publication’s analysis.
5. Canada. Canadian Human Rights Act. Sections 2, 3, 5 and 15. Sections 15(1)(a), 15(1)(g), 15(2) and 15(8) address employment and service justifications, accommodation and adverse effects. Undue hardship considers health, safety and cost. The applicable law and complaint route vary by sector and jurisdiction.
6. Canada. Accessible Canada Act. Sections 2, 5, 6, 7 and 69–71. The federal framework addresses barrier identification, removal and prevention. Sections 69–71 cover accessibility planning, consultation, feedback and reporting; they do not create an automatic entitlement to a particular AI tool or ISBN.
7. United Nations / UNICEF. Convention on the Rights of the Child: full text. Articles 2, 3, 5, 12, 13, 16, 17, 19, 23 and 31. The child’s own assisted expression is distinct from an adult publisher producing material for children. Best interests, evolving capacities, privacy and protection inform the rights discussed here.
8. UN Committee on the Rights of the Child. General Comment No. 25 (2021) on children’s rights in relation to the digital environment. CRC/C/GC/25, 2 March 2021. Guidance on digital inclusion, disability access, children’s participation, privacy and remedies. This publication recommends preserving effective ways to communicate without AI.
9. Canada. Privacy Act. Section 4 on collection of personal information.
10. Accessibility Standards Canada. CAN-ASC-6.2:2025 Accessible and Equitable Artificial Intelligence Systems. Voluntary standard. Agency, alternatives and accommodation effects.
11. Government of Canada. Common Core Document. 2020. Context on the domestic implementation of international human-rights obligations; not an independent source of individual remedies.
12. Edis. First-person account of building AccessSlice with Claude; AccessSlice public repository and creator documentation. The first-person account supports the Claude-assisted development narrative; the repository documents the software and tested printer support.
13. Luke Stoffel. A Machine That Holds All My Thoughts at Once. First-person account of AI and dyslexia.
14. Kirkus Reviews. Boy Refracted by Luke Stoffel. Publication corroboration, not verification of AI contribution.
15. Jayne Hayward, Beatrice Vincenzi, Christina Karpodini and Arthur Theil. How Do Disabled Visual Artists Use GenAI Tools for Business Administration Tasks. ASSETS 2025. DOI: 10.1145/3663547.3759708. Five qualitative interviews.
16. Innovation, Science and Economic Development Canada. Enhancing trust in artificial intelligence through increased transparency. Discussion Document, July 2026. Policy context for meaningful disclosure, differing risks and authority to act. Not enacted requirements; the consultation closed 23 September 2026.
17. Supreme Court of Canada. Moore v British Columbia (Education). 2012 SCC 61. Meaningful access to education and the role of disability supports; paragraph 33 sets out the initial discrimination test and the respondent’s burden of justification. The decision does not determine entitlement to an AI product.
18. Office of the Privacy Commissioner of Canada. Children’s Privacy Code. 20 October 2026. Further context on children’s participation, evolving capacities, accessible information and correction. Regulatory guidance interpreting PIPEDA, not a new Act; comments are open until 20 January 2027.
19. Supreme Court of Canada. British Columbia (Superintendent of Motor Vehicles) v British Columbia (Council of Human Rights), known as Grismer. [1999] 3 SCR 868. Individual assessment, accommodation and justification of a disability-related standard. Concerned driving eligibility, not AI access.
20. Supreme Court of Canada. Eldridge v British Columbia (Attorney General). [1997] 3 SCR 624. Effective communication and sign-language interpretation in publicly funded medical care; Charter application to implementing a government program. Not an AI-product entitlement.
21. Employment and Social Development Canada. Backgrounder 3: Artificial intelligence, accessibility and standards. Phase 2 digital-accessibility regulatory engagement. Checked 26 October 2026. Explains potential benefits, barriers and participation in development; supports discussion and does not set requirements.
22. Office of the Privacy Commissioner of Canada and privacy authorities of Quebec, British Columbia and Alberta. Joint Investigation of OpenAI OpCo, LLC. PIPEDA Findings #2026-002, 6 May 2026. Consent, accuracy of personal information, access, correction, deletion and accountability. Focused on GPT-3.5 and GPT-4; considered subsequent mitigation measures. Conclusions vary with the applicable federal and provincial laws.
Credits
Developed by ABILITYscaled through collaboration between Lance Lattrulo and Louis, his OpenAI-powered AI assistant. Lance directed the argument, editorial decisions and art direction. AI contributed research, drafting, revision, illustration generation and production. Final publication decisions rest with ABILITYscaled.
AI-generated illustration, developed through iterative human art direction.
