EU AI Act risk levels

The EU AI Act regulates AI by risk — some uses are banned, some are high-risk, some need transparency.

EU rules path · 1 of 4

What it is

The EU AI Act sorts AI uses into levels. Unacceptable-risk practices, such as social scoring, are banned. High-risk systems — for example in hiring, credit scoring or critical infrastructure — must meet strict requirements. Some systems, such as chatbots, carry transparency duties. Most other uses have no specific obligations beyond AI literacy.

Example

A tool that screens job applications is typically high-risk; a tool that drafts marketing copy usually is not.

What can go wrong

Misclassifying a system — and finding out late that it needed documentation, human oversight and registration.

How to secure it

List the AI systems you use or build, classify each by its intended purpose, and document the reasoning.

EU rules

Bans on unacceptable-risk practices apply since February 2025, and two more bans apply from December 2026. Fines for banned practices can reach €35 million or 7% of worldwide annual turnover, whichever is higher (lower caps apply to smaller companies). After the Digital Omnibus (July 2026), most high-risk obligations apply from December 2027, and from August 2028 for AI built into regulated products.

Plain-language summary, not legal advice.

Related service: AI Security