AI and accessibility: the European Accessibility Act and discrimination risks

Short answer: Accessibility touches AI in two ways. The European Accessibility Act (EAA, Directive (EU) 2019/882) requires many digital products and services to be usable by people with disabilities — including AI interfaces such as chatbots and voice assistants. In addition, you must prevent an AI system from discriminating against people with disabilities; the AI Act sets data and risk-management requirements to that end.

The European Accessibility Act

The EAA harmonises accessibility requirements for, among others, computers, smartphones, e-commerce, banking services, e-books and electronic communications. The directive is applicable from 28 June 2025.

For AI this means concretely that interfaces must be perceivable, operable and understandable. An AI chatbot in a webshop, a voice-controlled assistant or an automated customer portal falls within these requirements where it affects the relevant service.

The directive provides exemptions for micro-enterprises supplying services and a proportionality test where compliance would impose a disproportionate burden. That exemption is no free pass, however: it must be substantiated and documented, and does not apply automatically to large suppliers of AI services.

Discrimination risks of AI

A second, subtler risk is that an AI system unintentionally disadvantages people with disabilities. Speech recognition that fails to understand atypical speech, facial analysis that does not recognise certain faces, or a selection model that penalises atypical CVs — these are forms of indirect exclusion.

The AI Act addresses this through quality requirements for training data and mandatory risk management for high-risk systems, such as AI in recruitment or access to services. The aim is to detect and limit bias that leads to discrimination.

On top of this, general equal-treatment law continues to apply in full. An AI system that systematically disadvantages people with disabilities can constitute prohibited discrimination, regardless of whether the system complies with the AI Act. Compliance with the AI Act therefore does not rule out a discrimination claim.

Two frameworks, one design task

The EAA and the AI Act come together in design. Accessibility is not a separate check at the end, but a requirement on the system itself: representative data, tested interfaces and alternative interaction modes. This connects with the broader protection of vulnerable groups described under AI and minors.

What to do

• Check whether the EAA applies: do you supply consumer-facing digital products or services?
• Test interfaces for accessibility: do chatbots, voice and forms meet perceivable, operable, understandable?
• Check training data for representativeness of people with disabilities.
• Offer alternatives: an accessible route alongside the AI route.
• Document choices: record accessibility and bias tests, also with a view to transparency.

Accessibility and non-discrimination reinforce each other: designing inclusively also reduces legal risk.