EU AI Act Risk Categories

The EU AI Act does not treat all AI the same. A spam filter and a system that decides who gets a job carry very different risks, so the law sets rules according to the level of risk. At the top are practices considered so harmful that they are banned outright. Next are high-risk systems, which can be used only if they meet strict requirements. Below them are systems that must be transparent, such as chatbots and generators of synthetic content. Everything else falls into a minimal-risk category with no specific obligations. Understanding which category an AI system falls into is the first step in working out what the law requires.

General-purpose AI models, such as large language models, are regulated separately, with their own obligations and an extra tier for models that pose systemic risk. Categories also depend on how an AI system is used, not just the technology, so the same underlying model can be minimal risk in one application and high risk in another. For legal, compliance, risk and product teams, classifying AI systems correctly drives compliance planning, budgets and product decisions.

Definition of EU AI Act Risk Categories

The EU AI Act risk categories are the classifications under Regulation (EU) 2024/1689 that determine which obligations apply to an AI system, based on its intended purpose and potential to harm health, safety or fundamental rights, ranging from prohibited practices to minimal-risk systems.

They form the core structure of the EU AI Act, are assessed during an AI risk assessment and sit alongside separate rules for general-purpose AI models.

The Four Categories

Unacceptable risk covers prohibited practices, such as manipulative techniques that cause significant harm, exploiting vulnerabilities due to age or disability, social scoring, predicting criminality based solely on profiling, untargeted scraping of facial images, emotion recognition in workplaces and schools, certain biometric categorization and most real-time remote biometric identification in public spaces for law enforcement. The 2026 omnibus added AI that generates non-consensual intimate imagery or child sexual abuse material.

High risk covers AI systems that are safety components of products regulated under EU product safety law, listed in Annex I, and AI systems used in sensitive areas listed in Annex III.

Limited risk, often described this way although the law speaks of transparency obligations, covers chatbots, emotion recognition and biometric categorization systems, and generators of synthetic audio, image, video or text, which must be disclosed or marked.

Minimal risk covers all other AI systems, such as spam filters or inventory forecasting, which face no specific obligations under the Act, although voluntary codes are encouraged.

High-Risk Areas in Annex III

Annex III lists areas including biometrics, critical infrastructure, education and vocational training, and employment and worker management.

It also covers access to essential private and public services, such as credit scoring, risk assessment and pricing in life and health insurance, eligibility for public benefits and emergency call triage.

It includes law enforcement, migration, asylum and border control, and the administration of justice and democratic processes.

An Annex III system may fall outside the high-risk category if it does not pose a significant risk, for example because it performs a narrow procedural task, but systems that profile individuals are always treated as high risk.

Importance of EU AI Act Risk Categories in AI Governance

Risk categories determine obligations. Whether a system is prohibited, high risk, subject to transparency or minimal risk decides what the organization must do and how much it will cost.

They shape product design. Organizations can sometimes design or limit uses to avoid high-risk classification, or decide not to pursue prohibited uses at all.

They also set a model for internal governance. Many organizations use similar tiers in their own AI risk frameworks, even outside the EU.

EU AI Act Risk Categories and Financial Management

High-risk classification brings significant costs for documentation, testing, quality management, conformity assessment, monitoring and human oversight.

Prohibited practices carry the highest fines, up to 35 million euros or 7% of worldwide annual turnover, whichever is higher.

Early classification lets organizations budget accurately and plan compliance against deadlines, now set at December 2027 for Annex III systems and August 2028 for Annex I systems.

Classification also informs whether a product remains commercially viable in the EU once compliance costs are included.

Common Challenges

Classification depends on intended purpose and context, which can be hard to pin down for flexible AI tools. A general assistant may become high risk if it is used to screen job applicants.

The exemption for Annex III systems that do not pose significant risk requires careful assessment and documentation.

Guidance and standards continue to develop, so classifications may need to be revisited. Keeping a record of the reasoning makes it easier to update classifications later.