Article 9: Risk Management System for High-Risk AI
EU AI Act Article 9: providers of high-risk AI systems must run a continuous risk management system. What it covers, when it applies, what breaches cost and where to start.
Providers of high-risk AI systems placed on the EU market must run a risk management system across the system’s lifecycle (Article 9). It identifies, evaluates and reduces risks.
The rule applies from 2 December 2027 to Annex III systems (listed high-risk uses). For Annex I products, it applies from 2 August 2028 (Article 113(c)), except Section B (Article 2(2)). Provider fines reach EUR 15 million or 3% of worldwide annual turnover, whichever is higher (Article 99(4)). Next step: confirm that your system is high-risk under Article 6, then start a risk register.
Article 9 requires you to identify, evaluate, reduce and test risks
Identify risks
List the known and reasonably foreseeable risks to health, safety or fundamental rights when the system is used as intended (Article 9(2)(a)). Only risks you can reasonably reduce through design, development or technical information count (Article 9(3)).
Estimate and evaluate risks
Include risks under reasonably foreseeable misuse and risks that appear in post-market monitoring data (Article 9(2)(b) and (c)).
Reduce risks in a set order
First eliminate or reduce risks through design and development. Then add mitigation and control measures. Then give deployers information and, where appropriate, training. The remaining risk must be acceptable (Article 9(5)).
Test against defined metrics
Test during development and before placing the system on the market, against metrics and probabilistic thresholds defined in advance (Article 9(6) and (8)). Testing in real-world conditions is possible under Article 60 (Article 9(7)).
Consider children and vulnerable groups
Consider whether the system is likely to harm people under 18 and, where appropriate, other vulnerable groups (Article 9(9)).
Four engineering practices that produce Article 9 evidence
Article 9 prescribes no tools. These four practices are suggestions for a documented, repeatable process.
Risk register as code
Keep a machine-readable risk register (JSON or YAML) in version control next to the code. Record each risk, its likelihood and severity, the mitigation and the owner.
Risk checks in CI/CD
Add scans and tests to the pipeline that flag risk-relevant patterns, such as missing input validation or missing confidence thresholds.
Tests for known risks
Write tests that check each mitigation works, against metrics and thresholds defined in advance (Article 9(8)).
Monitoring after release
Feed production data into the register: performance drops, drift, odd outputs and human overrides (Article 9(2)(c)).
Risk management runs through the whole lifecycle
Article 9(2) calls the system a continuous iterative process run throughout the lifecycle, with regular systematic review and updating.
Identify and rate risks before each release.
Implement mitigations and test them.
Monitor after release and feed the findings back.
Common questions about Article 9
Does Article 9 apply to all AI systems?
No. Article 9 applies to high-risk AI systems, which Article 6 defines through Annex III and Annex I. It applies from 2 December 2027 to Annex III systems and from 2 August 2028 to Annex I systems (Article 113(c)). For high-risk systems related to products under Annex I Section B laws, only Articles 6(1), 60a and 102 to 112 apply, and Articles 57 to 59 apply only in so far as the requirements for high-risk AI systems have been integrated in that legislation (Article 2(2)).
What documentation does Article 9 require?
Article 9(1) requires the risk management system to be documented, and Annex IV point 5 requires a detailed description of it in the technical documentation. The Act lists no fixed set of documents. Record the risks, the measures, the judgement on the remaining risk and the test results.
How often must the risk management system be reviewed?
Regularly and systematically throughout the lifecycle (Article 9(2)). The Act sets no fixed interval.
Who is responsible for the risk management system?
The provider. Providers must ensure their high-risk AI systems meet the requirements of Section 2, Article 9 included (Article 16(a)). Deployers do not run it, but they must monitor the operation of the system and inform providers where relevant (Article 26(5)).
What fine applies if a provider breaks Article 9?
Fines for breaking the provider duties in Article 16 reach EUR 15 million or 3% of worldwide annual turnover, whichever is higher (Article 99(4)). For SMEs and small mid-cap companies, the lower of the two amounts is the limit (Article 99(6) and (6a)).
Read next: logging, oversight and technical documentation
Article 12: record-keeping
Covers the automatic event logs that support monitoring after release.
Article 14: human oversight
Covers the oversight measures that can reduce risks.
Article 11: technical documentation
Covers the documentation that must describe your risk management system (Annex IV, point 5).
Scanara prepares evidence from your repositories
Scanara reads your connected GitHub repositories and runs rule-based scans for standard AI and machine-learning patterns. It can miss non-standard code. Separately, it generates draft documentation from templates and your assessment answers as evidence for your compliance work. You make the classification decisions.