AI Trend Research

← AI glossary

AI glossary · Beginner · Safety & alignment

EU AI Act

The EU AI Act is an artificial intelligence regulation created by the European Union (EU). It sets different obligations depending on how much risk an AI system could pose to people.

About this term
Level
Beginner · Core ideas for first-time readers
Category
Basics
Added
2026-10-08
In plain words It is similar to how cars, toys and medicines each have different safety standards. The more dangerous a product, the more thoroughly it must be inspected before release. AI is different, though: one model can be used for many purposes, so it is hard to assign a level just by looking at the type, as you would with a physical product.

How it works

The EU AI Act uses a "risk-based approach." It divides AI into risk levels based on where it is used and sets different rules for each level. Uses whose risk is unacceptably high, such as covertly manipulating people or assigning social scores, are banned outright. "High-risk" uses that strongly affect people's lives, such as hiring, education, healthcare and critical infrastructure, carry strict obligations such as risk management, data quality management, record-keeping and human oversight. Where risk is limited, as with chatbots or AI-generated content, the key requirement is a transparency obligation to "disclose that it is AI," and most other AI is minimal risk and not specifically regulated.

Regulation by risk levelAI system → Risk classification; Risk classification → Prohibited; Risk classification → High risk; Risk classification → Limited risk; Risk classification → Minimal risk; High risk → Strict obligations; Limited risk → Transparency obligationAI systemWhere is it used?RiskclassificationProhibitedManipulating people,social scoringHigh riskHiring, education,healthcareLimited riskChatbots, generatedcontentMinimal riskMost AIStrict obligationsRisk management, humanoversightTransparencyobligationDisclose that it is AI
Regulation by risk level AI is assigned a risk level based on where it is used. The higher the level, the stricter the regulation.

Why it matters

As a leading example of a regulation that broadly covers AI as a whole, other countries and companies use it as a reference point. Any company that offers or uses AI services in the EU must comply with the law, even if the company is based outside the EU. As a result, global AI companies sometimes change product features or policies to suit the EU market. Companies that build general-purpose AI models used for many purposes (foundation models) also now have separate obligations such as documentation and information disclosure.

Things to know

Different provisions start applying at different times, so the law does not take effect all at once but in stages. Detailed criteria, such as what counts as high-risk, continue to be specified through guidelines and standards. Violating the law can bring large fines proportional to a company's revenue. There are both criticisms that the regulation slows innovation and views that it is necessary to protect citizens.

Example

This site also covered the news "OpenAI applies watermarks to ChatGPT and Codex generated text in response to EU rules." It is a case of a service provider responding to the transparency requirement in EU rules, which asks that AI-generated content be identifiable, by adding a marker to its outputs. In this way, the EU AI Act directly affects the product design of companies outside Europe too.

Where it came up on this site

Related terms

AI watermark Foundation model Sovereign AI Guardrails Alignment