Accessibility and the EU AI Act

Posted on Wednesday, 11 November 2026 by Henny Swan in Standards, Strategy

The EU Artificial Intelligence Act (EU AI Act) is already in force, with more of its rules coming into effect each year until 2028. So what does it mean if you build AI or use third-party AI tools, and where does accessibility fit in?

The Act is a framework for ensuring AI is safe and respects people’s fundamental rights. It promotes better governance of AI, and transparency when it’s used. If you’re building or using artificial intelligence in your consumer products in the EU, then you are subject to the Act.

The overlap between the Act and accessibility

The AI Act makes compliance with the Web Accessibility Directive and the European Accessibility Act (EAA) a legal requirement for providers of high-risk AI from December 2027. In practice, this means products and services meeting EN 301 549 and the Web Content Accessibility Guidelines (WCAG) 2.2.

The Act also requires that people are told when they’re interacting with AI or AI-generated content, and that this information is accessible. This applies to more than high-risk AI.

Who the Act affects

If you provide or deploy an AI tool in the EU, you’re subject to the Act, including if you’re based outside the EU but your customers are in the EU.

A provider is an organisation that develops and markets AI systems. This can be anything from chatbots and virtual assistants to screening tools and AI image generators.

A deployer is an organisation that integrates these third-party AI tools. This could be a financial institution using AI to predict monthly spending or a retailer using an AI chatbot to answer customer queries.

Risk levels and what they mean for you

The Act sets out duties that companies providing or deploying AI must meet categorised into four risk levels: unacceptable, high, limited, or minimal risk.

Each risk level has different duties. However, there is a common requirement at each level: organisations must take steps to help staff build AI literacy so they understand AI well enough to use it safely.

Unacceptable risk

This is AI that threatens people’s safety or rights, for example social scoring, emotion recognition in the workplace, or AI that exploits people’s age, disability, or circumstances. From December 2026, it also includes AI that generates non-consensual intimate images or child sexual abuse material (CSAM).

The requirement here is clear-cut. The Act states that this type of AI must not be used.

High risk

This is AI that makes or supports significant decisions about people, for example recruitment screening, university admissions, credit scoring, and benefit eligibility.

Most of the Act’s duties sit here. If you build high-risk AI, you must manage its risks, make sure people can oversee it, prove it complies, and make it accessible. If you deploy high-risk AI, you must use it as intended, monitor it, and tell people when it’s used.

Limited risk

This is AI that interacts with people or creates content, for example chatbots, virtual assistants, image generators, and writing tools.

You must tell people when they’re interacting with AI, and label AI-generated content as AI. The labels must also be accessible.

Minimal risk

This covers AI that poses little or no risk to people. This includes spam filters, AI in video games, route planning, and product recommendations.

There are no specific duties under the Act, but voluntary codes of conduct are encouraged.

Practical steps you can take

If you are using AI, here are some steps you can take to get ready for the Act:

Review how you are using AI and assess what products and services are subject to the Act
Embed accessibility into AI governance and ensure it is part of your corporate governance
Carry out assessments on your products against WCAG and EN 301 549 so you can evidence accessibility and document a roadmap to meet compliance
Carry out usability testing on your products with disabled people to validate how usable the products are
Ensure that any third-party AI tools are accessible and your procurement process assesses accessibility compliance
Document accessibility work so you can evidence meeting your obligations should you be challenged
Put in place a process for labelling uses of AI so people know when they are consuming AI-generated content

Summary

The Act doesn't replace accessibility law, it reinforces it. Whatever your sector, if you build or use AI, accessibility needs to be part of how you choose, build, and govern it.

What this means in practice depends on how you use AI:

Retail: if you use AI chatbots or virtual assistants, you must tell people they're interacting with AI, and make sure both the chatbot and that label are accessible
Financial services: if you use AI for credit scoring or life and health insurance pricing, it's high-risk, so you must monitor it, explain decisions, and carry out a fundamental rights impact assessment
Software as a Service (SaaS): if you build AI into your product, you're a provider, and if it's used for high-risk purposes like recruitment screening, you must make it accessible
Public sector: if you use AI to decide benefit eligibility or access to services, it's high-risk, and you must carry out a fundamental rights impact assessment
Any organisation using AI to recruit: AI that screens candidates is high-risk, so check your suppliers can evidence accessibility before you buy

The good news is that if you're already working towards making your product accessible in line with the European Accessibility Act (EAA), you're in good shape as EN 301 549 and WCAG, apply.

Next steps

If you’re not sure where to start, our Consultancy services can help you get a handle on where you are and what you need to do.

Have a look at our European Accessibility Act (EAA) FAQ for more answers to common questions about the Act itself and what’s covered.