EU AI Act

The EU has two laws that matter for accessible AI. The European Accessibility Act sets accessibility duties for many products and services. The EU AI Act regulates AI according to how risky its use is. They are different instruments with different goals, and a single AI product can fall under both.

Key points

The EU has two relevant laws, the European Accessibility Act for accessibility duties and the EU AI Act for risk-based AI rules, and one product can fall under both.
The EAA applies from 28 June 2025, and conformance is shown against EN 301 549, which incorporates WCAG 2.1 AA.
The AI Act sorts uses by risk, with the heaviest duties on high-risk uses like employment and essential services phasing in through 2026 and 2027.
The AI Act references accessibility aligned with the EAA, so a high-risk system is expected to be accessible, not on a separate track.
Outside the EU the picture is similar, with the ADA, Section 508, and Canada’s CAN-ASC-6.2 all reaching AI.

The European Accessibility Act

The European Accessibility Act, or EAA, is Directive (EU) 2019/882. Obligations on businesses apply from 28 June 2025. It covers a broad range of consumer products and services, such as e-commerce, banking, e-books, and transport ticketing. Conformance is usually demonstrated against the European standard EN 301 549, which incorporates WCAG 2.1 level AA.

Many AI-powered services sit inside this scope. An AI assistant on a banking site or a chatbot on an online shop is not outside accessibility law just because it is AI. If the service is covered, the AI parts of it have to be accessible too. See WCAG applicability.

The EU AI Act

The EU AI Act is Regulation (EU) 2024/1689. It came into force on 1 August 2024 and its rules phase in over the following years. It is a product-safety style regulation built around risk, so the duties on a system depend on how it is used. The Act sorts uses into broad groups.

Prohibited practices. A small set of uses, such as certain kinds of social scoring, is banned outright. These rules apply first, from early 2025.
High-risk systems. Uses such as employment, education, and access to essential services carry the heaviest duties, around risk management, data quality, transparency, human oversight, and accessibility. These obligations phase in later, through 2026 and 2027.
General-purpose AI. Providers of general-purpose models carry their own transparency duties, supported by a code of practice, applying from 2025.
Limited-risk uses. Some uses, such as a chatbot, mainly carry a duty to tell people they are dealing with an AI system rather than a human.

Where the two laws meet

The AI Act references accessibility requirements that align with the EAA, so a high-risk AI system is expected to be accessible as part of meeting the Act. Accessibility is not a separate track you can postpone. The European Disability Forum keeps an updated guide to monitoring the AI Act from a disability-rights point of view, tracking the guidance the European Commission issues on prohibited uses, high-risk duties, and general-purpose models.

What advocates are watching

Disability organizations are not only reacting to the law, they are working to shape it and to build the skills to use AI well. The European Disability Forum runs a project called Empowered by AI that teaches AI literacy to disabled people across Europe, so they can take part in how these systems are built and governed rather than only receiving them.

Beyond the EU

The wider picture is similar. In the United States, the Americans with Disabilities Act and Section 508 both lean on WCAG, and the US Access Board has held hearings on AI and disability. Canada has published CAN-ASC-6.2:2025, a standard for accessible and equitable AI systems. The common thread is that accessibility duties now reach AI, and being an AI product is not treated as an excuse.

Audience: Product Manager

Two dates anchor planning for the EU.

EAA obligations are live from 28 June 2025 for covered consumer services.
AI Act duties phase in through 2027, with the heaviest load on high-risk uses such as hiring and essential services.

Work out early whether your product is a covered service under the EAA and whether any use is high-risk under the AI Act, because that decides which obligations apply. For procurement, expect to provide an accessibility conformance report, and remember that the two laws stack rather than substitute.

Audience: Accessibility Specialist

Be exact about the chain of references. The EAA points at EN 301 549, which incorporates WCAG 2.1 AA, and the AI Act in turn references accessibility aligned with the EAA. So for a high-risk system the practical accessibility baseline still flows back to WCAG, plus the Act’s own duties on transparency, human oversight, and documentation. Track which version of EN 301 549 is current, since that determines the exact WCAG version in force.

Audience: Engineer

Two AI Act duties land in the build.

Transparency means disclosing when a user is interacting with an AI system, and that disclosure has to be accessible like any other content.
Human oversight means there has to be a real way for a person to review, correct, or stop the system, which overlaps directly with the Operable requirements for stop and undo.

Design those hooks in rather than bolting them on.