The Article 9 lifecycle
Article 9(1) is explicit: the risk-management system is "a continuous iterative process planned and run throughout the entire lifecycle of the high-risk AI system, requiring regular systematic review and updating." That has four practical implications:
- You cannot front-load Article 9 into a pre-market document and walk away. Updates are mandatory.
- Significant changes during the operational phase — new training data, model updates, deployment to new contexts — re-trigger the analysis.
- The post-market monitoring under Article 72 is a feeder into Article 9, not a parallel process.
- Documentation must show the iteration, not just the latest snapshot.
The five-step procedure under Article 9(2)
- Identification and analysis of known and reasonably foreseeable risks. Risks to health, safety, fundamental rights — including discrimination. The "reasonably foreseeable" qualifier extends to misuse: if a deployer might foreseeably use the system in a way that produces a risk, that misuse-risk is in scope.
- Estimation and evaluation of risks emerging when the system is used in accordance with its intended purpose.
- Estimation and evaluation of other possibly arising risks under reasonably foreseeable misuse.
- Adoption of appropriate and targeted risk-management measures. Designed to mitigate the risks identified. Article 9(3) requires reduction "as far as possible through adequate design and development."
- Evaluation of residual risk. Article 9(4) requires that residual risks are judged acceptable, and communicated to the deployer. The acceptability judgment is documented and signed off by an accountable person.
Where Article 9 intersects with other articles
- Article 10 (data governance). Bias and representativeness risks are identified in Article 9, addressed via the Article 10 data-governance procedures.
- Article 14 (human oversight). Where a residual risk cannot be reduced by design, Article 14 oversight measures may bring the residual risk to acceptable levels. Article 14 is a mitigation tool inside Article 9.
- Article 15 (robustness). Accuracy, robustness, and cybersecurity targets fall out of Article 9 — they are not standalone numbers but the Article 9 conclusions about acceptable performance.
- Article 72 (post-market monitoring). The plan under Article 72 is the operating mode of the Article 9 system in production.
- Article 73 (incident reporting). Serious incidents feed back into Article 9 risk re-evaluation.
Integration with existing standards
- ISO/IEC 42001:2023 (AI management system). The Annex A controls map well to Article 9; ISO 42001 is the closest fit for an integrated framework.
- ISO 31000 (general risk management). The vocabulary and process steps are compatible; ISO 31000 alone does not satisfy AI Act content requirements but is a sensible chassis.
- ISO 14971 (medical-device risk management). For Annex I medical-device AI, ISO 14971 must be extended to cover Article 9's misuse-risk and fundamental-rights elements; the integrated risk file is a single document.
- NIST AI Risk Management Framework. Useful structure for governance and measurement but does not satisfy Article 9 alone — too high-level.
Article 9(8) — minors
Where the system is likely to be accessed by or impact persons under 18, Article 9(8) requires explicit consideration of the impact on those persons. This is operative in education AI (most of Annex III §3), in some healthcare AI, and in any consumer system reasonably accessible to children. The risk register must include child-specific considerations: developmental impact, reduced capacity to recognise AI, parental notice, age-appropriate transparency.
What "good" looks like
- A live risk register with version history, not a static PDF.
- Linkages from each Article 9 risk to the specific Article 10/14/15 mitigation that addresses it.
- Sign-off on residual-risk acceptability by an accountable person, not a working-group consensus.
- Evidence of post-market monitoring data feeding back: at least one risk-register update driven by Article 72 input.
- For Annex III §3/§5/§7 (where minors or vulnerable groups feature heavily), explicit Article 9(8) and Article 9(2)(b) misuse analyses.