Filed under — EU AI Act · Annex III · High-Risk AI · Risk Classification · Compliance
Prefer this source on Google →EU AI Act Annex III: Complete High-Risk AI Systems List
Annex III lists the uses that make an AI system high-risk under the EU AI Act; Annex I products are the other route. Learn which use cases trigger obligations before the 2 December 2027 deadline.
Free check · no account
Check your own system against what this article describes.
EU AI Act Annex III: Complete High-Risk AI Systems List
Updated 2 October 2026 — correction. This guide said a system had to meet an Annex III condition and an EU product-law condition to be high-risk, listed practices that Article 5 prohibits as high-risk examples, and put general-purpose AI under Article 50. It now says that Article 6 gives two separate routes (Annex I products, or Annex III uses, with the Article 6(3) exception), marks the prohibited practices as prohibited, places general-purpose AI in Chapter V (Articles 51 to 56), completes the Annex III list from the Official Journal text, and no longer names products or offers a paid check.
Updated 4 August 2026 — deadline changed. The Digital Omnibus (adopted by the European Parliament on 16 June 2026 and the Council on 29 June 2026) deferred the Annex III high-risk obligations from 2 August 2026 to 2 December 2027. The Article 50 transparency obligations were not deferred and have applied since 2 August 2026. This article has been corrected accordingly.
The EU AI Act divides AI systems into four risk categories: unacceptable, high, limited, and minimal. Only high-risk AI systems face the full weight of compliance obligations — Articles 9 through 15, technical documentation, conformity assessment, CE marking, and post-market monitoring.
Whether your AI system is high-risk is decided by Article 6, which gives two separate routes. Under Article 6(1), an AI system is high-risk if it is a product, or a safety component of a product, covered by the EU product legislation listed in Annex I and that product must undergo a third-party conformity assessment. Under Article 6(2), an AI system is high-risk if its use is listed in Annex III, a legally binding list of use cases — unless the Article 6(3) exception applies. If your system is on neither route, it is not high-risk, though it may still face the transparency requirements of Article 50 (Regulation (EU) 2024/1689).
Annex III high-risk obligations apply from 2 December 2027. The statutory maximum for breaching the high-risk obligations is €15 million or 3% of global annual turnover under Art. 99(4) — whichever is higher, except for SMEs and start-ups, where Art. 99(6) caps the fine at whichever is lower. The €35M / 7% tier applies only to the prohibited practices in Article 5. This article provides the complete Annex III list, explains what each category covers, and shows how to work out whether your system is high-risk.
How Annex III Works
Annex III is not a static list. Under Article 7, the European Commission can amend it by delegated act to add or modify use cases within the areas it already lists; it cannot add a new area that way. The list as published in the Official Journal covers eight areas.
There are two separate routes to high-risk status. Either one is enough; neither needs the other:
- Annex III route (Article 6(2)): the system's intended use falls under one of the eight Annex III areas (biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, or justice and democratic processes). By the Article 6(3) exception, such a system is not high-risk where it does not pose a significant risk of harm to health, safety or fundamental rights — for example when it only performs a narrow procedural task or a preparatory task — but it is always high-risk if it profiles natural persons.
- Annex I route (Article 6(1)): the system is a product, or a safety component of a product, covered by the EU harmonisation legislation listed in Annex I (e.g., medical devices, machinery, toys), and that product must undergo a third-party conformity assessment under that legislation.
If your system falls into no Annex III area and does not reach the regime through Annex I, it is not high-risk under the EU AI Act, even if it poses significant ethical or social risks. The Act is use-case-specific, not capability-specific.
The Complete Annex III List
Here are all eight Annex III areas, with explanations and examples.
1. Biometrics (Annex III.1)
AI systems used for remote biometric identification, biometric categorisation or emotion recognition, in so far as their use is permitted under Union or national law.
| Subcategory | Description | Examples |
|---|---|---|
| Remote biometric identification | Identifying people by comparing their biometric data (face, gait, voice) with a reference database; verification that only confirms a person is who they claim to be is excluded | Facial recognition at airports, police surveillance cameras, stadium entry systems |
| Biometric categorisation | Assigning people to categories according to sensitive or protected attributes or characteristics inferred from their biometric data | Categorisation that Article 5 does not prohibit (see below) |
| Emotion recognition | Identifying or inferring people's emotions from their biometric data | Emotion recognition outside the workplace and education |
Key point: Not all biometric systems are high-risk. Biometric authentication (unlocking your phone with your face) is not covered by Annex III.1, which expressly excludes verification that only confirms a person's identity. And some biometric uses are not high-risk but prohibited under Article 5: categorising individuals by their biometric data to deduce or infer race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation (Article 5(1)(g)), and inferring emotions in the workplace or in education institutions, except for medical or safety reasons (Article 5(1)(f)). Those carry the €35M / 7% tier.
2. Critical Infrastructure (Annex III.2)
AI systems used as safety components in the management and operation of critical digital infrastructure, road traffic, or the supply of water, gas, heating or electricity.
| Subcategory | Description | Examples |
|---|---|---|
| Critical digital infrastructure | AI that is a safety component in managing or operating critical digital infrastructure | Safety systems in data-centre or network operation |
| Road traffic | AI that is a safety component in managing or operating road traffic | Traffic signal and traffic management systems |
| Water, gas, heating, electricity supply | AI managing supply, demand, or safety in utility networks | Smart grid optimization, predictive maintenance for power plants, water treatment control systems |
Key point: The system must be a safety component. An AI that optimizes energy costs is not high-risk; an AI that prevents blackouts or pipeline failures is. AI in vehicles themselves reaches the regime through Annex I product law, not through Annex III.2.
3. Education and Vocational Training (Annex III.3)
AI systems used to determine access to education, evaluate learning, assess the level of education a person will receive, or monitor students during tests.
| Subcategory | Description | Examples |
|---|---|---|
| Admission and enrollment | AI that decides who gets accepted to schools, universities, or training programs, or assigns people to them | University admissions algorithms, scholarship award systems |
| Assessment and evaluation | AI that evaluates learning outcomes, including when the outcomes are used to steer a person's learning process | Automated essay grading, plagiarism detection that affects grades |
| Level of education | AI that assesses the appropriate level of education a person will receive or be able to access | Placement tests that decide which course or track a student can access |
| Monitoring during tests | AI that monitors and detects prohibited behaviour of students during tests | AI proctoring systems that flag students for cheating |
Key point: The trigger is access, evaluation, placement or test monitoring. A tutoring tool that evaluates learning outcomes to steer a student's learning can fall under Annex III.3(b), even if it does not affect grades or admissions.
4. Employment, Worker Management, and Self-Employment (Annex III.4)
AI systems used in recruitment, hiring, promotion, termination, task allocation, or monitoring of workers.
| Subcategory | Description | Examples |
|---|---|---|
| Recruitment and hiring | AI that places targeted job advertisements, analyses and filters applications, or evaluates candidates | Video interview analysis, résumé parsing and ranking tools |
| Promotion and termination | AI that makes decisions affecting terms of work, promotion, or termination | Performance review algorithms, layoff selection models |
| Task allocation and monitoring | AI that allocates tasks based on individual behaviour or personal traits, or monitors and evaluates workers' performance and behaviour | Warehouse task assignment, driver monitoring and ratings, call center performance scoring |
Key point: This is one of the broadest high-risk areas. If your AI touches hiring, firing, or worker evaluation, it very likely falls under Annex III.4; whether the Article 6(3) exception applies has to be assessed and documented.
5. Essential Private and Public Services (Annex III.5)
AI systems used to evaluate eligibility for or grant access to essential services and benefits.
| Subcategory | Description | Examples |
|---|---|---|
| Public benefits eligibility | AI used by or for public authorities to evaluate eligibility for essential public assistance benefits and services, including healthcare, or to grant, reduce, revoke or reclaim them | Eligibility screening for public housing or welfare |
| Creditworthiness and credit scoring | AI that evaluates the creditworthiness of natural persons or establishes their credit score; systems used to detect financial fraud are excluded | Credit scoring models, loan approval algorithms, buy-now-pay-later eligibility checks |
| Life and health insurance | AI used for risk assessment and pricing of natural persons in life and health insurance | Underwriting and premium-setting models |
| Emergency services dispatch | AI that evaluates and classifies emergency calls, dispatches or prioritises emergency first response, or triages emergency healthcare patients | Emergency call triage systems, ambulance dispatch optimization |
Key point: The system must affect access. An AI that helps you compare loan offers is not high-risk; an AI that decides whether you get approved is.
6. Law Enforcement (Annex III.6)
AI systems used by or on behalf of law enforcement authorities, or by EU bodies in their support, in so far as their use is permitted under Union or national law.
| Subcategory | Description | Examples |
|---|---|---|
| Victim risk assessment | AI that assesses the risk of a natural person becoming the victim of criminal offences | Risk assessment in domestic violence cases |
| Polygraph and lie detection | AI used as a polygraph or similar tool | AI-powered lie detectors, voice stress analysis |
| Evidence evaluation | AI that evaluates the reliability of evidence in the investigation or prosecution of criminal offences | DNA match probability, forensic image analysis |
| Risk of offending | AI that assesses the risk of a person offending or re-offending not solely on the basis of profiling, or assesses personality traits or past criminal behaviour | Recidivism risk tools |
| Profiling in investigations | AI used for profiling natural persons in the detection, investigation or prosecution of criminal offences | Criminal network analysis |
Key point: This category applies only to law enforcement use, including use on behalf of law enforcement authorities. Predicting that a person will commit a crime based solely on profiling or on personality traits is not high-risk but prohibited (Article 5(1)(d)). A private company's fraud detection is not covered by Annex III.6, and Annex III.5(b) expressly excludes fraud detection from the credit-scoring category.
7. Migration, Asylum, and Border Control (Annex III.7)
AI systems used by or on behalf of competent public authorities to manage migration, asylum applications, or border security.
| Subcategory | Description | Examples |
|---|---|---|
| Polygraphs | AI used as a polygraph or similar tool | Lie detection at borders |
| Risk assessment | AI that assesses a security, irregular-migration or health risk posed by a person who intends to enter or has entered a Member State | Risk profiling for border checks |
| Visa and asylum applications | AI that assists in examining applications for asylum, visas or residence permits, and associated complaints, including assessing the reliability of evidence | Visa risk assessment tools, automated review of asylum appeal documents |
| Identification | AI that detects, recognises or identifies people; the verification of travel documents is excluded | Biometric identification at borders |
Key point: This category is narrow and applies to competent public authorities and EU bodies managing migration, asylum and borders.
8. Administration of Justice and Democratic Processes (Annex III.8)
AI systems used to assist judicial authorities or to influence elections and referenda.
| Subcategory | Description | Examples |
|---|---|---|
| Administration of justice | AI used by or on behalf of a judicial authority to research and interpret facts and the law and to apply the law to a set of facts, or used in a similar way in alternative dispute resolution | AI-assisted legal research and sentencing support for courts |
| Democratic processes | AI intended to influence the outcome of an election or referendum or people's voting behaviour; tools whose output people are not directly exposed to, such as campaign logistics, are excluded | Voter targeting algorithms, AI-generated political ads |
Key point: AI used by or on behalf of a judicial authority to assist in sentencing or case research is high-risk. AI used by lawyers for their own research is not listed in Annex III.8 (unless it is used on behalf of a judicial authority or in alternative dispute resolution).
What If Your System Spans Multiple Categories?
If your AI system falls into more than one Annex III area, record every point that applies. For example:
- An AI system that screens job applicants (Annex III.4) and uses remote biometric identification (Annex III.1) is listed under both points.
- An AI system that assesses creditworthiness (Annex III.5) and assesses a person's risk of offending for law enforcement (Annex III.6) is listed under both points.
You cannot "choose" the easier category. The requirements of Articles 9 to 15 are the same for every high-risk system, but some duties depend on the point: biometric systems under point 1 may need a notified body for conformity assessment (Article 43(1)), and deployers of credit-scoring and insurance systems under points 5(b) and (c) must carry out a fundamental rights impact assessment (Article 27).
What If Your System Is Not on the List?
If your AI system falls into no Annex III area and does not reach the regime through Annex I, it is not high-risk under the EU AI Act. You may still face other obligations:
- Article 50 (Transparency): Providers of systems that interact directly with people must ensure people are informed they are interacting with AI; providers of systems that generate synthetic content must mark it as such; deployers of emotion recognition or biometric categorisation systems must inform the people exposed; deployers of deepfakes must disclose them.
- Chapter V (General-Purpose AI, Articles 51 to 56): If you provide a general-purpose AI model, you face separate obligations under Article 53, and under Article 55 if the model has systemic risk.
Many AI systems — recommendation engines, content moderation, marketing optimization, internal analytics — are not listed in Annex III and are not high-risk unless they reach the regime through Annex I.
Common Misclassifications
Several classification errors recur:
- Overclaiming high-risk status: Providers assume their system is high-risk because it uses sensitive data or makes important decisions. The EU AI Act is use-case-specific, not risk-based in the general sense. If your system is neither in Annex III nor an Annex I product or safety component, it is not high-risk.
- Underclaiming high-risk status: Providers assume their system is not high-risk because it "only assists" humans. If the system influences hiring, credit access, or law enforcement decisions, a human making the final call does not on its own take it out of Annex III. The Article 6(3) exception covers only systems that do not pose a significant risk — for example narrow procedural or preparatory tasks — it never covers profiling, and a provider who relies on it must document the assessment (Article 6(4)) and register the system (Article 49(2)).
- Ignoring edge cases: A system used for internal HR analytics may not be high-risk. The same system used to make promotion decisions is listed (Annex III.4(b)).
The test is the system's intended purpose, use case, and deployment context.
How to Determine If Your System Is High-Risk
Follow this decision tree:
-
Is your system a product, or a safety component of a product, covered by the EU legislation listed in Annex I, and does that product need a third-party conformity assessment?
- Yes → Your system is high-risk under Article 6(1). Example: your AI is a safety component of a medical device that needs a notified body under the Medical Device Regulation.
- No → Continue to step 2.
-
Does your system's intended purpose fall under one of the Annex III areas?
- Example: Your system uses facial recognition, but only to unlock a phone (verification, not identification). → Not listed.
- Example: Your system uses facial recognition to identify individuals in a crowd. → Listed (Annex III.1(a)).
- Example: Your AI optimizes commercial ad targeting. → Not listed (and not a safety component under Annex I).
- Not listed → Your system is not high-risk. Check Article 50 for transparency obligations.
- Listed → Continue to step 3.
-
Does the Article 6(3) exception apply?
- If the system profiles natural persons → High-risk, always.
- If it does not pose a significant risk of harm — for example it only performs a narrow procedural or preparatory task — it is not high-risk, but you must document that assessment before placing it on the market and register it (Articles 6(4) and 49(2)).
- Otherwise → High-risk under Article 6(2).
A system that is high-risk by either route must comply with Articles 9–15, technical documentation, conformity assessment, and post-market monitoring.
Timeline: When Annex III Becomes Enforceable
| Date | Milestone |
|---|---|
| 2 August 2025 | GPAI obligations and the penalty regime apply |
| 2 August 2026 | Article 50 transparency obligations apply (not deferred) |
| 2 December 2027 | Annex III high-risk obligations apply |
| 2 August 2028 | Annex I high-risk obligations apply (AI in regulated products) |
Annex III high-risk obligations become legally binding on 2 December 2027, and penalties apply immediately after that date. Note that if your system also interacts with people directly, the Article 50 transparency obligations already bind you today — they were not deferred.
Final Checklist: Is Your System High-Risk?
Use this checklist to assess your system:
- My system is, or is a safety component of, a product under the Annex I legislation that needs a third-party conformity assessment
- My system falls into at least one Annex III area (biometrics, infrastructure, education, employment, essential services, law enforcement, migration, justice and democratic processes)
- My system is used for the specific purpose listed in that area (not a tangential use case)
- I have checked whether the Article 6(3) exception applies, and whether the system profiles natural persons
- I have documented the risk classification with legal justification
- If high-risk, I have begun implementing Articles 9–15 obligations (risk management, data governance, technical documentation, record-keeping, transparency, human oversight, accuracy, robustness and cybersecurity)
If you checked the first box, your system is high-risk under Article 6(1). If you checked the second and third boxes, it is high-risk under Article 6(2) unless the Article 6(3) exception applies, and that exception never covers profiling. If you checked neither the first box nor the second, your system is likely not high-risk; record why.
This article is for informational purposes only and does not constitute legal advice. Consult a qualified EU AI Act attorney for guidance specific to your situation.
Related dispatches