Disability perspective on Regulating Artificial Intelligence

The Disability perspective on Regulating Artificial Intelligence is the European Disability Forum's (EDF) foundational position paper on AI regulation, published on 2 November 2021 in direct response to the European Commission's proposed Artificial Intelligence Act (April 2021). EDF is the independent pan-European umbrella organisation representing the interests of 50 million persons with disabilities across the EU, with member organisations from every EU and EEA country and European-level NGOs representing specific disability types. The paper is EDF's authoritative statement on what rights-respecting AI regulation must include from a disability perspective, grounded in the UN Convention on the Rights of Persons with Disabilities (UNCRPD). Its organising principle is "Nothing About Us Without Us" — the disability-rights proposition that persons with disabilities must be active participants, not passive subjects, in any policy that shapes their lives — applied to the emerging field of AI governance.

Five requirements for disability-inclusive AI regulation

The position paper sets out five requirements for the proposed regulation:

Accessibility. All AI-based technologies must be accessible to persons with disabilities by design. Accessibility is framed not as an optional retrofit but as a precondition for any AI system's legitimate use in contexts where it interacts with people — a design requirement that enters at development, not at deployment review.

Protection from harm. AI regulation must carry explicit safeguards against discrimination, privacy violations, and data-protection risks that fall specifically on persons with disabilities: systems that assess people on disability-related characteristics, and those that compound existing exclusion in employment, education, and access to services.

Strong governance. Human rights impact assessments, accessible feedback channels, and accessible complaint and redress mechanisms must be mandatory for AI systems that affect fundamental rights. The accessibility of governance infrastructure is itself a substantive requirement: a redress mechanism that a blind user cannot navigate is no redress.

Global standards. AI systems developed under European rules must satisfy those standards wherever they are deployed. A regulatory framework that permits export of non-compliant systems would undermine its own rights-protection logic and produce a two-tier world where European persons with disabilities are protected and others are not.

Meaningful inclusion. Persons with disabilities and their representative organisations must be structurally mandated in AI policy development, technical standards-setting, and development teams — not offered consultation windows after substantive choices have been made. The "Nothing About Us Without Us" principle demands participation, not comment.

Red lines: prohibited AI uses

The position paper named categories of AI use that must be absolutely prohibited:

Employment screening systems that discriminate on disability. The EDF cited Austria's Public Employment Service (AMS) algorithm as the documented case of an AI system that systematically disadvantaged persons with disabilities in job-matching, applying disability status as a negative scoring signal in an automated gatekeeping tool. This case became a flagship example in EDF's EU AI Act advocacy of what a "high-risk AI system" must be prevented from doing.

Emotion-recognition systems in employment and education. Video-interviewing AI platforms that score candidates on facial expressions, speech cadence, and appearance penalise the non-standard presentation that characterises many deaf, blind, deafblind, autistic, and otherwise disabled people. EDF advocacy materials specifically cite the HireVue case — the AI-video-interviewing platform challenged by the US Federal Trade Commission in 2021 — as evidence that without prohibition these systems filter out qualified disabled candidates before any human interviewer is involved.

AI-based gatekeeping for social rights and benefits. Automated systems that determine eligibility for healthcare, disability support, or social benefits without meaningful human review are characterised by EDF as surveillance rather than service delivery: they concentrate decisional power over disabled people's basic economic security in algorithmic systems that are opaque to the people they affect and practically difficult to appeal. The paper frames this as an application of the UNCRPD right to independent living and inclusion in the community (Article 19) — a right that cannot be operationalised through an AI whose decisions cannot be interrogated.

Real-time biometric identification and categorisation in public spaces. This includes systems capable of inferring disability status, emotional state, or health characteristics from publicly-observable cues. EDF argues such systems compound the already-heightened social visibility navigated by visibly disabled people and that permitting them under AI regulation would be a regressive step relative to existing anti-discrimination law.

EU AI Act process and advocacy outcomes

EDF engaged continuously in the EU AI Act legislative process from the 2021 Commission proposal through the trilogue negotiations that produced the final text (entered into force 1 August 2024). The 2023 Board Resolution on the EU AI Act, adopted on 1 April 2023 during the active trilogue phase, sharpened the 2021 position paper into specific legislative demands: expansion of the prohibited AI practices list, stronger obligations for high-risk AI systems affecting disabled people's rights, and mandatory disability-inclusive conformity assessment procedures.

The final AI Act's accessibility requirements for high-risk AI systems were described by EDF as a "partial win on accessibility" — an acknowledgment that the legislation incorporated disability-access obligations into the high-risk AI framework while falling short of EDF's full demands on prohibited practices and structural participation requirements. The framing signals continued implementation advocacy rather than a settled legislative win.

Disability Inclusive AI project and implementation phase

Following the AI Act's adoption, EDF launched the five-year Disability Inclusive Artificial Intelligence project (2023–2027) to translate its advocacy into member-state implementation monitoring. The project's AI Policy Officer, Kave Noori — who joined EDF in January 2023 — became EDF's primary specialist in AI and disability rights, presenting to European Parliament committees on Internal Market and Civil Liberties and Justice matters, and participating in the UN Conference of State Parties to the CRPD. In October 2024 EDF published the Disability-Inclusive Artificial Intelligence Act: A Guide to Monitor Implementation in Your Country — a practical toolkit for disability organisations and national equality bodies — with an updated edition released in October 2025.

Position in the corpus

This publication is the corpus's first entry in the disability-rights movement area on AI — the Synthesizer flagged zero disability-axis entries when queuing it. The EDF position paper fills the slot of a rights-based, disability-led advocacy publication on AI regulation: a document whose authority derives from EDF's representative mandate across 50 member-country umbrella organisations and European disability-type NGOs, and which deploys the UNCRPD framework as the legal foundation for its AI governance demands rather than grounding them in novel AI-specific rights claims.

The disability-rights movement's engagement with AI regulation sits at a specific intersection of the make-AI-good movement: not primarily a "make AI safer" argument in the existential or large-model sense, but a "make AI non-discriminatory and accessible" argument rooted in existing human-rights law and in the concrete documented harms of deployed systems. EDF's use of UNCRPD Articles 4, 9, and 19 as the legal frame for AI accountability — rather than treating AI as a problem requiring new rights — is the distinctive register of this strand, adjacent to but structurally separate from the algorithmic-accountability litigation strand (AlgorithmWatch, Foxglove) and the AI-safety-treaty strand (Stop Killer Robots, PauseAI). The corpus had no publication anchoring the disability-rights strand of the make-AI-good movement before this entry.