Introduction

The AI Act is the European regulation governing the use of artificial intelligence (AI). It classifies AI systems according to the risk they pose to people, in order to ensure safe and ethical use. This classification is the starting point of any compliance effort: our guide to the AI Act for businesses then details the timeline and the obligations of each stakeholder.

The regulation entered into force on 1 August 2024 and applies in stages. At the end of 2026, the prohibitions and transparency rules are already applicable; the obligations for high-risk systems have been postponed to 2 December 2027 by the Digital Omnibus. So that your teams can recognise these levels in their day-to-day work, our AI training courses for businesses cover this topic.

The different AI risk levels under the AI Act

The AI Act framework is based on a risk-based approach, divided into four levels: unacceptable, high, limited and minimal risk.

Unacceptable risk

AI systems classified as posing an unacceptable risk have been prohibited since 2 February 2025. This includes technologies that manipulate human behaviour in harmful ways or that exploit the vulnerabilities of specific individuals or groups, the social scoring of people, emotion recognition in the workplace or at school, and the building of facial recognition databases by scraping images from the internet. Two further prohibitions are added on 2 December 2026: systems designed to generate non-consensual intimate content or child sexual abuse material.

High risk

High-risk AI systems are subject to strict requirements. This covers applications such as recruitment, credit scoring of individuals or critical infrastructure. The provider must in particular manage risks, document the system and provide for human oversight; the company that uses it must monitor how it operates. These obligations will apply on 2 December 2027 for the Annex III areas, and on 2 August 2028 for AI embedded in products that are already regulated. To learn more about successfully integrating AI, read our AI Roadmap.

Limited risk

Limited-risk AI systems must meet transparency obligations, applicable since 2 August 2026. For example, chatbots must inform users that they are interacting with a machine, and deepfakes must be flagged. This category does not require compliance as strict as high risk, but still calls for precautions. For more information on automating customer service with AI, read our article on AI Chatbots.

Minimal risk

Minimal-risk AI systems pose little risk to users' rights and do not require any particular compliance measures. This includes the majority of AI applications used today, such as spam filters or content recommendations. Companies only need to take steps to ensure that their teams are proficient with these tools (Article 4).

Why risk classification matters for businesses

AI risk classification is crucial for businesses because it allows them to understand the regulatory obligations that apply to their AI systems. By identifying the risk level of their applications, companies can make sure they comply with legal and ethical standards, thereby avoiding penalties and protecting their reputation. To take action, follow our AI Act compliance checklist for SMEs.

For a full assessment of your AI maturity level, read our guide on the AI Audit.

💡 Did you know? diagonall is a French AI agency that supports companies in their AI transformation with training, strategic consulting and custom solutions.

Conclusion

The AI Act provides a clear framework for managing the risks associated with the use of AI. By classifying AI systems according to their risk level, it allows companies to know where to focus their efforts and to ensure that their AI practices comply with safety and ethical standards.

Frequently asked questions

What is the AI Act?

The AI Act is the European regulation (EU) 2024/1689 that governs the use of artificial intelligence by classifying AI systems according to their risk level. It applies in stages between 2025 and 2028.

What risk levels does the AI Act define?

The AI Act defines four risk levels: unacceptable, high, limited and minimal, each with its own specific compliance requirements.

When do the high-risk obligations apply?

On 2 December 2027 for the Annex III areas (recruitment, credit, education, etc.) and on 2 August 2028 for AI embedded in regulated products, following the postponement decided by the Digital Omnibus in 2026.

Why is AI risk classification important?

It allows companies to comply with legal and ethical standards, avoid penalties and protect their reputation by ensuring the safe use of AI.

Ready to take action?

Let's talk AI

Book a demo