High-Risk AI Systems: The Complete List Under EU AI Act
High-Risk AI Systems: The Complete List Under EU AI Act
The EU AI Act classifies AI systems into four risk levels. High-risk systems face the strictest requirements — technical documentation, audit logging, human oversight, conformity assessment, and more.
Annex III of the EU AI Act defines exactly which AI systems are legally classified as high-risk. Here is the complete list with practical examples.
Annex III: The 8 High-Risk Categories
1. Biometric Identification and Categorisation
AI systems used to:
Real-world examples: Facial recognition login, biometric employee attendance, emotion detection in video interviews
Exception: Real-time remote biometric identification in public spaces is prohibited entirely (Annex I), not just high-risk.
---
2. Critical Infrastructure Management
AI systems used as safety components in:
Real-world examples: Predictive maintenance in power grids, autonomous traffic signal control, anomaly detection in water treatment
---
3. Education and Vocational Training
AI systems that:
Real-world examples: AI-powered university admission scoring, exam monitoring software (proctoring), adaptive learning systems that grade students
---
4. Employment, Worker Management, and Access to Self-Employment
AI systems used for:
Real-world examples: Applicant tracking systems with AI scoring, AI-powered interview analysis, Uber/Deliveroo driver assignment algorithms
This is the category most relevant to HR Tech companies in Germany.
---
5. Access to Essential Private and Public Services
AI systems that determine or significantly influence access to:
Real-world examples: AI credit scoring used by banks, automated welfare eligibility assessment, AI-powered insurance underwriting
---
6. Law Enforcement
AI systems used by police and law enforcement for:
Real-world examples: Predictive policing tools, lie detection systems, facial recognition for criminal suspect identification
---
7. Migration, Asylum, and Border Control
AI systems used in:
Real-world examples: Automated border control gates, AI visa processing, traveller behaviour analysis
---
8. Administration of Justice and Democratic Processes
AI systems that assist courts in:
Real-world examples: AI-powered legal research assistants used by judges, political microtargeting systems
---
Am I Affected?
You are likely affected if you sell software used in HR, healthcare, banking, education, or critical infrastructure — especially if AI is involved in any decision that affects people.
The test is not what you call your product. The EU AI Act looks at what your system does, not what your marketing says. An "AI assistant" that ranks job candidates is treated as a high-risk employment AI.
What High-Risk Classification Means in Practice
If your system is high-risk, before August 2, 2026 you must have: