AI Act Risk Classification

The legal framework

The AI Act (Regulation (EU) 2024/1689) attaches obligations to the risk class of a system. Classification is therefore the first legal question, and the answer depends on the intended purpose of the system and the context in which it is used, not on the technology as such.

Prohibited practices

Article 5 prohibits, among others, manipulative or exploitative techniques that cause significant harm, social scoring, untargeted scraping of facial images, emotion recognition in the workplace and education (except for medical or safety reasons) and certain uses of biometric identification. Regulation (EU) 2026/1744 adds two prohibitions on AI systems that generate non-consensual intimate imagery or child sexual abuse material; they apply from 2 December 2026.

High-risk systems

Article 6 gives two routes. First, an AI system that is a safety component of a product, or is itself a product, covered by the EU harmonisation legislation listed in Annex I (machinery, medical devices, toys and others) and that requires third-party conformity assessment. Second, a system listed in Annex III, which covers eight areas:

• biometrics;
• critical infrastructure;
• education and vocational training;
• employment and worker management;
• access to essential private and public services, including credit scoring and life and health insurance pricing;
• law enforcement;
• migration, asylum and border control;
• administration of justice and democratic processes.

An Annex III system is not high-risk if, under Article 6(3), it does not pose a significant risk of harm to health, safety or fundamental rights, for example because it only performs a narrow procedural task, improves the result of a previously completed human activity, detects deviations from earlier decision patterns, or performs a preparatory task. A system that performs profiling of natural persons is always high-risk. A provider that relies on the exemption must document its assessment before placing the system on the market and register the system.

Transparency duties

Article 50 requires that people are told when they interact with an AI system, that synthetic audio, image, video or text is marked as artificially generated, that deployers of emotion recognition or biometric categorisation inform the persons exposed, and that deepfakes are disclosed. These duties can apply to a system on top of any other class.

The remainder

Systems that fall in none of these categories, such as spam filters or AI in video games, carry no specific obligations, but Article 4 on AI literacy still applies.

In practice

A bank uses an AI model to assess consumer credit applications. Creditworthiness assessment of natural persons is listed in Annex III, point 5, and the model builds a profile of the applicant, so the Article 6(3) exemption is not available. The system is high-risk.

The same bank deploys a chatbot for general questions. It is not in Annex III, so it is not high-risk, but Article 50 requires that customers are told they are talking to an AI system.

A logistics company uses an AI tool that sorts incoming invoices into folders. The tool performs a purely preparatory task with no decision about individuals, and in any case it is not in an Annex III area. It carries no specific obligations other than AI literacy.

A recruitment agency uses software that scores CVs. This falls under employment in Annex III and involves profiling, so it is high-risk. See AI in HR and recruitment.

Each conclusion should be written down with the reasoning. Where a high-risk system processes personal data, a data protection impact assessment is often required as well.

A useful way to work is to keep a short classification note per system: the intended purpose, the Annex III area (if any), the reasoning on Article 6(3), the transparency duties that apply and the name of the person who approved the conclusion. When the system or its use changes, the note is revisited.

Frequently asked questions

When do the high-risk rules apply?

Following Regulation (EU) 2026/1744, from 2 December 2027 for Annex III systems and from 2 August 2028 for Annex I systems. Prohibitions and transparency duties are not postponed. See the timeline.

Can the Commission change the lists?

Yes. The Commission can amend Annex III by delegated act and issue guidelines on classification, so the lists should be monitored.

What if I am unsure about the class?

Make a documented assessment of intended purpose, area of use and effect on individuals. Where doubt remains, treat the system as high-risk until a lawyer has confirmed otherwise.

Does the class depend on the model or on the use?

On the use. The same underlying model can be minimal-risk in one application and high-risk in another.

How this connects

Our AI Act register records each classification together with the articles it relies on, and our AI Act compliance service has an attorney confirm it. Once a system is high-risk, the next question is your role, covered in provider or deployer.