For about two years, a date kept turning up in coverage of artificial intelligence and disability rights: 2 August 2026. That was when the European Union’s AI Act was supposed to start applying to the systems that decide things about people. Benefits eligibility. Job applications. School admissions. Credit. Insurance pricing. Which emergency call gets dispatched first.
That date has now passed, and most of those rules did not start.
On 27 July 2026 a set of amendments known as the AI Omnibus entered into force, published in the Official Journal three days earlier as Regulation (EU) 2026/1744. Among other changes, it moved the compliance dates for high-risk AI systems. Stand-alone systems of the kind listed in Annex III of the Act, which is where benefits and hiring sit, now apply from 2 December 2027. High-risk AI built into regulated physical products, such as machinery and lifts, applies from 2 August 2028.
If you read something last year telling you that European law would require human oversight of benefit-decision algorithms this summer, that article is now out of date. Here is what replaced it.
What still applies right now
The delay is not a repeal, and a fair amount of the AI Act is already live.
The prohibitions have applied since 2 February 2025. These are the practices the Act bans outright rather than regulates, and one of them names disability directly. Article 5(1)(b) of the Act as adopted, Regulation (EU) 2024/1689, prohibits placing on the market or using an AI system that “exploits any of the vulnerabilities of a natural person or a specific group of persons due to their age, disability or a specific social or economic situation” where the effect is to materially distort their behaviour in a way likely to cause significant harm. That is in force today, across the EU, and enforcement of it started on 2 August 2026. The Omnibus amended Article 5 to add the prohibitions described below, and the Commission notes that its published text of the article does not yet show those amendments.
Also banned since February 2025: AI that infers emotions in workplaces and schools, except for medical or safety reasons, and biometric categorization systems that sort people by protected characteristics.
Rules for general-purpose AI models and the governance structures behind the Act took effect on 2 August 2025. The transparency obligations in Article 50, which cover telling people when they are interacting with an AI system or looking at AI-generated content, started on 2 August 2026 alongside national and EU enforcement powers.
Two further dates are close. From 2 December 2026, a new prohibition applies to AI systems that generate non-consensual sexual imagery and child sexual abuse material. The same date is the transition deadline for certain providers whose synthetic-content systems were already on the market before August 2026.
What actually moved, and why it matters to this audience
The delayed category is worth reading closely, because Annex III is not an abstract list. It is a fairly precise description of the decisions that shape a life with a disability.
Annex III point 5(a) covers AI systems used by or for public authorities “to evaluate the eligibility of natural persons for essential public assistance benefits and services, including healthcare services, as well as to grant, reduce, revoke, or reclaim such benefits and services.” Point 5(d) covers systems that triage emergency calls and prioritize dispatch. Point 4 covers recruitment, candidate filtering, and decisions about promotion, termination and task allocation at work. Point 3 covers admission to education and the assessment of learning outcomes. Points 5(b) and 5(c) cover credit scoring and risk pricing for life and health insurance.
The obligations that attach to those systems are the practical ones: risk management, data governance and bias testing, technical documentation, logging, human oversight, and accuracy requirements. Those are the duties that now begin in December 2027 rather than August 2026.
What disability organizations have said
The European Disability Forum, the umbrella body representing organizations of people with disabilities across Europe, published a briefing and a summary article in June 2026 setting out its objections. Its position is that the Omnibus “rolls back many of the fundamental rights safeguards established by the Artificial Intelligence Act,” and it lists six specific concerns.
Three of them are worth naming here because they are concrete. First, the Omnibus lets developers and deployers process sensitive personal data, including health and disability information, without individual consent for the purpose of detecting and correcting bias. EDF notes this applies not only to high-risk systems but also, under certain conditions, to all AI systems, and says the change “significantly broadens access to highly sensitive data.” Second, equality bodies lose the ability to request documentation directly from AI providers and must now go through market surveillance authorities instead, which EDF says “risks weakening the ability of specialised authorities to effectively investigate and respond to discrimination.” Third, the AI literacy duty changed from a requirement that organizations ensure staff competence to a requirement that they support it. EDF’s example is pointed: “in decisions about disability-related benefits, insufficient scrutiny of AI outputs could lead to unfair decisions that are difficult to challenge.”
EDF also called the new prohibition on non-consensual intimate imagery “welcome,” while objecting to the process that produced it.
The Commission’s own framing is different. It describes the package as “a targeted simplification of the AI rulebook while preserving strong safeguards for people’s safety and fundamental rights.”
Where this leaves readers in Canada
None of this is Canadian law, and Canada has no equivalent statute in force.
The Artificial Intelligence and Data Act was part of Bill C-27, the Digital Charter Implementation Act, 2022. It was introduced in June 2022, passed second reading in April 2023, and never made it out of the Standing Committee on Industry and Technology. The parliamentary session ended on 6 January 2025 and the bill died with it. Parliament’s own record now lists C-27 under “Historical information.”
What does exist federally is narrower and often overlooked. The Treasury Board’s Directive on Automated Decision-Making governs how federal institutions use automated decision systems, and it requires an Algorithmic Impact Assessment. Amendments that took effect on 25 April 2023 extended the directive to internal services such as hiring and security screening, required the Algorithmic Impact Assessment to be published before a system launches, and, in the Chief Information Officer’s own words, “introduce new questions into the AIA concerning the reasons for automation and potential impacts on persons with disabilities.”
That is a policy instrument covering the federal public service, not a law covering banks, insurers, landlords or provincial disability programs. Provincial income support, home care allocation and disability tax determinations sit outside it.
What to do with this information
Two things are useful to keep straight.
If you are following EU developments because they tend to set the pace for the rest of us, the timeline to watch is now December 2027 for the systems that decide about benefits and jobs, not August 2026. Anything you read that still gives the older date needs checking against the Commission’s published implementation timeline, which is updated and public.
If you are dealing with an automated decision in Canada right now, the AI Act is not your route. Human rights legislation, the duty to accommodate, and the appeal processes attached to whichever program made the decision are what apply. If the decision came from a federal institution, the published Algorithmic Impact Assessment for that system is a public document worth asking for.
We have written before about how automated decisions produce disability bias and what rights already apply, and about what happens when an algorithm decides your care hours.
This is general information about published law and policy, not legal advice. If an automated decision has affected your benefits, employment or care, a community legal clinic or an advocacy organization in your province can tell you what applies to your situation.
Sources
- European Commission, AI Omnibus enters into force, 27 July 2026
- European Commission AI Act Service Desk, Timeline for the Implementation of the EU AI Act
- European Commission AI Act Service Desk, Article 5: Prohibited AI practices
- European Commission AI Act Service Desk, Annex III
- European Disability Forum, 6 ways the AI Omnibus weakens protections for persons with disabilities, 23 June 2026
- Parliament of Canada, LEGISinfo: C-27, 44th Parliament, 1st session
- Treasury Board of Canada Secretariat, Amendments to the Directive on Automated Decision-Making
