Start with system scope
Determine whether the application is an AI system, which output is used and who acts as provider or deployer.
Back to Annex III overview
EU AI Act guidance 2026
Commission guidelines for high-risk AI classification
The draft guidelines help organisations assess Article 6, Annex III and high-risk obligations in the right order.
Connect Article 6 and Annex III
Assess whether the intended purpose falls under Article 6(2) and a specific Annex III point.
Document the Article 6(3) assessment
Explain why the exception does or does not apply, including profiling, rights impact and human oversight.
Turn classification into obligations
Connect the outcome to risk management, data governance, documentation, logging, transparency, oversight and FRIA where needed.
Status and planning
From draft to implementation planning
Treat this guidance as current, but not final. For decisions with legal consequences, the formal legal text remains leading.
Draft guidelines published
The Commission published the high-risk classification guidelines with separate downloads for general principles, Annex I and Annex III.
Feedback period closes
The Commission asks for input on clarity and usefulness of the examples before the final version is adopted.
Planning for certain Annex III areas
After the political agreement on the AI Omnibus, the Commission refers to this date for certain high-risk areas including biometrics, infrastructure, education, work, migration, asylum and border control.
Planning for product-related high-risk AI
For AI systems integrated into products, such as robotics or industrial machinery, the Commission refers to this later date.
Annex III areas
All high-risk domains in Annex III
The classification route runs by area. Use the domain pages to connect the examples from the guidelines to the internal AI inventory.
Annex III point 13 routes
Biometrics
The 19 May 2026 draft guidelines cover point 1(a), 1(b) and 1(c), with strong attention to context, purpose, consent, real-time use and exceptions.
View domain pageAnnex III point 23 routes
Critical infrastructure
The Commission splits this area into critical digital infrastructure, road traffic and the supply of water, gas, heating and electricity.
View domain pageAnnex III point 34 routes
Education and vocational training
The four routes are access/admission/assignment, evaluation of learning outcomes, assessment of education level and monitoring or detecting prohibited student behaviour.
View domain pageAnnex III point 42 routes
Employment and worker management
The HR route is already expanded in the theme pages for recruitment/selection and worker management/monitoring.
View domain pageAnnex III point 54 routes
Essential private and public services and benefits
The routes are essential public assistance benefits and services, creditworthiness/credit score, life and health insurance risk assessment/pricing, and emergency first response dispatch or triage.
View domain pageAnnex III point 63 routes
Law enforcement
The draft guidelines cover victim risk, polygraphs, evidence reliability, offending/reoffending and profiling in detection, investigation or prosecution.
View domain pageAnnex III point 73 routes
Migration, asylum and border control management
The routes are polygraphs, risk assessment for entry/stay, assistance with asylum/visa/residence applications and detecting/recognising/identifying persons.
View domain pageAnnex III point 82 routes
Administration of justice and democratic processes
The guidelines cover assistance to judicial authorities/ADR and systems that can influence elections or voting behaviour.
View domain pageEvidence and obligations
Connect classification to the evidence file
A high-risk conclusion is only useful when it connects to concrete obligations. These articles form the bridge from classification to execution.
First route check for your AI system
Use the classifier to structure the intended purpose, Annex III area, Article 6(3) assessment and first evidence items.
Open the classifierFrequently asked questions
Short answers for classification, evidence and next steps under Annex III.
What changes with the draft Commission guidelines of 19 May 2026?
The draft guidelines provide practical explanation for Article 6 and Annex III of the AI Act. They clarify how providers and deployers should assess whether a system is an AI system, whether it falls within an Annex III area and which examples are or are not treated as high-risk.
Are the guidelines final already?
No. The Commission published the draft guidelines on 19 May 2026 and asks for feedback until 23 June 2026. Organisations can use the guidance as an assessment framework now, but should update their files when the final version is published.
Which date should I use for Annex III high-risk AI planning?
For planning, the Commission refers after the political agreement on the AI Omnibus to 2 December 2027 for certain Annex III high-risk areas and 2 August 2028 for product-related high-risk AI systems. Always check formal legal text and the latest Commission publications before decisions are finalised.
How should this page be used in AI governance work?
Use the page as a starting point for inventory, classification and evidence work. For each system, document the intended purpose, Annex III route, Article 6(3) assessment, provider and deployer roles and the relevant obligations under Articles 9 to 15, 26 and 27.