What are the EU AI Act risk categories?

EU AI Act risk categories are the four levels the European Union's AI Act uses to decide how strictly an AI system is regulated: unacceptable risk, which is banned; high risk, which carries strict obligations; limited risk, which carries transparency duties; and minimal risk, which carries no new rules under the Act.

Why it matters for your business

The category decides what the Act requires, and it follows what a system is used for, not the technology behind it. Classifying each system is an early step in AI governance. Over-classifying pays for controls nobody requires. Under-classifying risks fines of up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher, for banned practices, and up to EUR 15 million or 3% for most other breaches. For SMEs the lower of the two figures applies.

How it works

As of October 2026:

Unacceptable risk: banned. Practices judged a clear threat to people's safety, livelihoods and rights, such as social scoring and emotion recognition in workplaces. Most bans have applied since 2 February 2025. A further ban, on AI that generates non-consensual sexually explicit content, takes effect in December 2026. High risk: strict obligations. AI used as a safety component of regulated products, and AI used in listed sensitive areas such as employment, education and access to essential services. Duties include risk assessment, documentation, logging and human oversight. They apply from 2 December 2027 for the listed areas and from 2 August 2028 for products. Limited risk: transparency. People must be told when they are dealing with a chatbot, and AI-generated content such as deepfakes must be identifiable. These duties have applied since 2 August 2026. Content-generating systems already on the market have until 2 December 2026 to mark their output. Minimal risk: no new rules. The European Commission says the vast majority of AI systems fall here, spam filters among them.

Example

Picture a 150-person staffing agency that uses a spam filter, a website chatbot and software that ranks CVs. The first is minimal risk. The second is limited risk: visitors must be told they are talking to a machine. The third is likely to be high risk, because employment is a listed area. One company, three sets of duties. The scenario is illustrative.

Costs, limits and when not to use it

The four-level picture is a simplification.

General-purpose AI models sit outside it. The large models behind many AI tools have their own rules, in application since August 2025. A listed area does not always mean high risk. The Act exempts systems that perform only a narrow procedural or preparatory task, but never systems that profile individuals. The category is half the answer. Duties also depend on the company's role: building a system and using one carry different obligations. The dates have moved once. The Digital Omnibus on AI, in force since 27 July 2026, postponed the high-risk deadlines. Check the current position before planning.

This is general information, not legal advice.