EU AI Act Articles 9 and 15: risk management, accuracy, robustness and cybersecurity for high-risk AI

Why this page: once a system is classified as high-risk under Article 6, Articles 9 and 15 are where an engineering team feels the Act. They read like a safety and security engineering brief. This page quotes the Service Desk text, based on the EUR-Lex consolidated version as at 27 July 2026. It is not legal advice; see the EU AI Act explainer.

Article 9: risk management

Article 9(1) says a risk management system shall be established, implemented, documented and maintained in relation to high-risk AI systems. Paragraph 2 calls it a continuous iterative process planned and run throughout the entire lifecycle of a high-risk AI system, requiring regular systematic review and updating.

The steps include the identification and analysis of the known and the reasonably foreseeable risks that the high-risk AI system can pose to health, safety or fundamental rights and the estimation and evaluation of the risks that may emerge when the high-risk AI system is used in accordance with its intended purpose, and under conditions of reasonably foreseeable misuse. A further step uses data gathered from the post-market monitoring system referred to in Article 72.

Two sentences matter for assurance evidence. Paragraph 5 requires that the measures make the relevant residual risk associated with each hazard, as well as the overall residual risk of the high-risk AI systems acceptable. Paragraph 8 says testing shall be carried out against prior defined metrics and probabilistic thresholds that are appropriate to the intended purpose of the high-risk AI system. Paragraph 10 lets providers subject to other Union risk-management rules combine the processes: those aspects may be part of, or combined with, the risk management procedures established pursuant to that law.

Article 15: accuracy, robustness and cybersecurity

Article 15(1) requires that high-risk systems achieve an appropriate level of accuracy, robustness, and cybersecurity, and that they perform consistently in those respects throughout their lifecycle. Paragraph 3 adds that the levels of accuracy and the relevant accuracy metrics of high-risk AI systems shall be declared in the accompanying instructions of use.

On robustness, paragraph 4 says it may be achieved through technical redundancy solutions, which may include backup or fail-safe plans. Systems that keep learning must be built to address feedback loops.

Paragraph 5 names AI-specific attacks. Measures shall include, where appropriate, ways to prevent, detect, respond to, resolve and control for attacks trying to manipulate the training data set (data poisoning), or pre-trained components used in training (model poisoning), inputs designed to cause the AI model to make a mistake (adversarial examples or model evasion), confidentiality attacks or model flaws. The vocabulary lines up with NIST AI 100-2, though the Act does not cite it in the text we fetched.

Reading the two together (editorial judgment)

This section is the site’s judgment, not the Commission’s.

• Article 9 asks for a process, Article 15 for properties of the product. A supplier needs evidence of both: the hazard log and the test results against stated thresholds.
• “Declared accuracy metrics” make a vendor’s datasheet part of the compliance record. A buyer can ask for the declared metrics and the test set they came from.
• The fail-safe wording echoes aerospace practice. See runtime assurance and simplex architectures for how that pattern is described elsewhere on this site. That similarity is not a finding that either satisfies the other.

Dates

The Commission’s AI Act page says Annex I high-risk rules have an extended transition period until 2 August 2028 and Annex III use cases have been extended to 2 December 2027. Confirm dates against the current text.

Questions

Are there harmonised standards that give a presumption of conformity for Article 15?

Not on the pages we fetched, so this site makes no claim. Check the Official Journal for current references before treating any standard as a route to conformity.

Does Article 9 replace an existing safety process?

Paragraph 10 lets the Article 9 aspects be part of, or combined with the risk management procedures required under other Union law. It does not say an existing process automatically satisfies it.