The two routes into high-risk
A system becomes high-risk one of two ways under Article 6. Most teams only need to check the second.
Annex I, embedded safety components
Your AI is a safety component of a product, or is itself a product, already covered by EU harmonised safety law that requires third-party conformity assessment. Medical devices, motor vehicles, civil aviation, machinery, lifts, and marine equipment all sit here. The AI compliance work folds into the product certification you already run. These requirements apply from 2 August 2027.
Annex III, standalone systems
Your AI operates in one of eight listed socio-economic domains where a decision lands on a person. No underlying product regulation is needed. This is the route that catches ordinary software companies, and it became enforceable on 2 August 2026.
Sitting outside both lists means your system is not high-risk. It may still owe the Article 50 transparency duties, and it is still banned if it does anything in Article 5.
The eight Annex III domains
Read these as job descriptions rather than industries. What matters is what the system decides, not what sector you sell into.
Biometrics
Remote biometric identification where it remains permitted, biometric categorisation based on sensitive attributes, and emotion recognition outside the contexts Article 5 bans outright.
Critical infrastructure
AI as a safety component in the management and operation of road traffic, digital infrastructure, and the supply of water, gas, heating, and electricity.
Education and vocational training
Deciding admissions, assigning people to institutions, assessing learning outcomes, judging the appropriate level of education, and monitoring for prohibited behaviour during exams.
Employment and worker management
Targeting job adverts, filtering applications, evaluating candidates, promotion and termination decisions, task allocation based on personal traits, and monitoring performance and behaviour.
Access to essential services
Eligibility for public benefits, creditworthiness and credit scoring outside fraud detection, risk pricing in life and health insurance, and triage of emergency calls and dispatch.
Law enforcement
Assessing the reliability of evidence, polygraphs, individual offending and re-offending risk, and profiling people during detection, investigation, or prosecution.
Migration, asylum, and border control
Polygraphs, risk assessment of people entering a Member State, examining applications for asylum, visa, and residence permits, and detecting or identifying people outside travel document checks.
Justice and democratic processes
Assisting judicial authorities with researching and interpreting facts and law, applying law to facts, and systems intended to influence the outcome of an election or voting behaviour.
Points 4 and 5 catch the most software companies. If you filter job applicants, score creditworthiness, or price life and health insurance, you are building a high-risk system even if the model itself is a few hundred lines of scikit-learn.
The Article 6(3) derogation
Article 6(3) can pull a system back out of high-risk even when it sits inside an Annex III domain. It applies where the system does not pose a significant risk of harm to health, safety, or fundamental rights, which the Act narrows to four situations.
- The system performs a narrow procedural task.
- The system improves the result of a human activity that is already complete.
- The system detects decision patterns or deviations from prior patterns without replacing or influencing human review.
- The system performs preparatory work for an assessment relevant to an Annex III use case.
The profiling exception swallows most attempts to use this. Any system that profiles natural persons, meaning automated processing of personal data to evaluate things like work performance, economic situation, health, preferences, or behaviour, stays high-risk. No derogation is available.
Relying on Article 6(3) is not a quiet decision. You must document the assessment before placing the system on the market, and you still register it in the EU database.
Provider obligations
If you build the system and put it on the market under your own name, Chapter III Section 2 is your list.
- Risk management system running across the entire lifecycle, reviewed and updated as the system changes.
- Data governance covering training, validation, and testing sets, examined for bias, gaps, and representativeness.
- Technical documentation matching Annex IV, ready before the system goes to market.
- Automatic logging of events over the system lifetime, retained for at least six months.
- Instructions for use that let a deployer operate the system correctly and understand its limits.
- Human oversight designed into the system so an operator can interpret output, override it, and stop the system.
- Accuracy, robustness, and cybersecurity appropriate to the intended purpose, sustained across the lifecycle.
- Quality management system covering the whole thing, documented in writing.
- Conformity assessment, EU Declaration of Conformity, CE marking, and registration in the EU database before market placement.
- Post-market monitoring, plus serious incident reporting to the national authority.
Deployer obligations
Buying a high-risk system rather than building one gives you a shorter list, and it is mostly operational.
- Operate the system according to the provider instructions for use.
- Assign human oversight to people with the competence, training, authority, and support to intervene.
- Make sure input data is relevant and sufficiently representative for the intended purpose.
- Monitor operation, and suspend use and inform the provider when you spot a risk or a serious incident.
- Keep the automatically generated logs for at least six months.
- Tell workers and their representatives before putting a high-risk system into use in the workplace.
- Inform people when they are subject to a decision made or assisted by a high-risk system, and explain the role it played.
Article 25 can move you from deployer to provider. Put your name or trademark on a high-risk system, modify it substantially, or change its intended purpose so it becomes high-risk, and the full provider list above becomes yours. See what companies need to do for the role test.
Fundamental Rights Impact Assessment
Article 27 puts the FRIA on the deployer, before first use. It applies to public bodies and private entities providing public services, and to any deployer using high-risk AI for creditworthiness assessment or risk pricing in life and health insurance.
The assessment has to describe how you will use the system, over what period and how often, which categories of people it affects, the specific risks of harm to them, how human oversight will work, and what you will do if a risk materialises. You notify the market surveillance authority of the result.
Deadlines and penalties
2 August 2026
Annex III high-risk requirements became enforceable.
2 August 2027
Annex I embedded safety component requirements apply.
Breaching high-risk obligations carries fines up to 15 million euro or 3% of total worldwide annual turnover, whichever is higher. Authorities can also restrict, withdraw, or recall the system from the EU market.
Full schedule: EU AI Act implementation timeline
How Prismor helps with the EU AI Act
Most of the Act's operational load comes down to two questions: which AI systems do you run, and can you show what they did. Prismor answers both for the AI inside your codebase.
Find every AI system you run
Phasing out Article 5 practices and classifying under Annex III both start from an inventory. Prismor finds the SDKs, models, agent configs, and MCP servers in your repos, including the ones nobody wrote down.
Records that hold up in an audit
Article 12 requires high-risk systems to log their operation automatically. Prismor captures every tool call and file touch with the session context an auditor asks about.
Human oversight you can enforce
Article 14 expects a person who can step in. Prismor holds risky agent actions for approval and blocks the ones that break your policy.
The AI Posture Analyzer is free. Public repos scan instantly with no signup. Private repos sign in with GitHub and run on your own access.
Frequently Asked Questions
Two routes lead to high-risk. Annex I covers AI acting as a safety component in a product already subject to third-party conformity assessment under EU harmonised safety law, such as medical devices, vehicles, and machinery. Annex III lists eight standalone domains including biometrics, critical infrastructure, education, employment, essential services such as credit scoring, law enforcement, migration, and justice. If your system sits in neither, it is not high-risk.