Business guides – contracts, HR, AI & compliance

AI Act risk classification – when is an AI system high-risk?

Practical guide to AI Act risk classification: prohibited practices, high-risk according to Article 6 and Annex I/III, transparency according to Article 50, and how companies should document the screening.
Published Mallbutiken
AI Act risk classification – when is an AI system high-risk?

Risk classification under the EU AI Act is not about assigning a general "low, medium, or high" risk rating. The company must first determine the type of AI usage involved, the organization's role, and whether specific provisions apply to the system.

Facts verified as of October 1, 2026. This guide provides general information and does not replace individual legal advice.

The timeline is important: The AI Act became largely applicable on August 2, 2026. Following the AI Omnibus, the new schedule dictates that the rules for high-risk cases under Annex III will begin to apply on December 2, 2027, and for high-risk AI embedded in regulated products under Annex I on August 2, 2028. The transparency requirements in Article 50, however, have been in effect since August 2, 2026.

A practical screening model

Question What you check
Is the solution an AI system within the meaning of the regulation? Define technology, function, and intended purpose before proceeding.
Is there a prohibited use? Screen against Article 5.
Could it be high-risk? Article 6 together with Annex I or Annex III.
Do transparency requirements apply? Article 50, for example, certain interactive AI systems, synthetic content, deepfakes, and other specific uses.
What role does the organization have? Provider, deployer/user, importer, distributor, or other relevant role.
Is there other regulation? GDPR, labor law, confidentiality, consumer law, sector regulation, etc.

1. Is the AI use prohibited?

The first check should be whether the use is covered by a prohibition in Article 5. Prohibited uses are not the same as "high-risk": as a rule, prohibited AI may not be used in the manner described by the prohibition. Therefore, screening should be done before the business spends time on a general risk score.

Document which use case has been assessed, what functions the system actually has, and why it is deemed to fall within or outside of a prohibition. This is particularly important when AI is used for profiling, biometric functions, or decisions that affect people.

2. Is the system high-risk under Article 6?

The AI Act has two central paths to high-risk classification. The first concerns AI that is a safety component of, or is itself, certain products subject to harmonized EU product legislation and requires third-party assessment. The second concerns certain use cases in areas specified in Annex III.

Annex III covers particularly sensitive areas, including parts of biometrics, critical infrastructure, education, employment, access to certain essential services, law enforcement, migration, and the administration of justice. However, a company using AI in HR, for instance, does not automatically mean that every AI function is high-risk; the concrete function and the intended purpose must be analyzed.

The European Commission's current high-risk guidelines are still marked as drafts. They are therefore guidance and examples, not a replacement for the regulation text itself.

3. Do the transparency requirements in Article 50 apply?

A system does not need to be high-risk to be subject to transparency requirements. Article 50 has applied since August 2, 2026, and includes obligations for certain AI systems that interact directly with humans and rules on labeling or information for certain forms of AI-generated or manipulated content.

For deployers, specific information obligations may be triggered by emotion recognition, biometric categorization, deepfakes, and certain AI-generated texts on matters of public interest. Exceptions and details vary, so the specific use needs to be checked against Article 50 and the Commission's guidelines.

4. "Not high-risk" does not mean "no risk"

A common mistake is to end the analysis when a system does not fall under the high-risk provisions. A generative AI tool may still process personal data, contain confidential information, create incorrect material, impact customers, or be used in processes where human oversight is necessary.

The company should therefore keep legal classification under the AI Act and internal operational risk separate. An internal risk matrix can prioritize which controls are needed even when the system is not legally classified as high-risk.

The easiest way to keep these assessments together is an AI registry where each use case is linked to screening, responsibility, risk assessment, and decisions.

Document the classification

The AI Governance Package 2026 includes an AI registry, AI risk assessment, AI impact assessment, supplier review, AI policy, and governance workflow.

Price: 199 SEK.

See AI Governance Package 2026

AI registry · AI Governance step-by-step · AI policy for companies

Sources

Read also AI in drinking-water utilities – high-risk classification, safety testing and manual fallback. Assess the rules and responsibilities that apply in the specific case.

Templates in this article

Explore the topic