Accuracy, Robustness, and Cybersecurity

Article 15 requires high-risk AI systems to meet appropriate levels of accuracy, robustness, and cybersecurity throughout their operational lifecycle. These technical performance requirements ensure that high-risk AI systems deliver reliable outputs and are protected against adversarial manipulation.

Accuracy and Robustness

Article 15(1) requires that high-risk AI systems achieve appropriate levels of accuracy for their intended purpose and be as consistent as possible throughout their lifecycle. Article 15(3) addresses robustness: systems must be resilient to errors, faults, and inconsistencies that may occur during normal operation or as a result of interaction with other systems or environments. They must also handle edge cases and unusual inputs without producing outputs that are dangerous or harmful.

Cybersecurity

Article 15(5) requires high-risk AI systems to be resilient against attempts by unauthorised third parties to alter their use, outputs, or performance. Providers must implement appropriate technical and organisational measures to protect against model poisoning, adversarial examples, data poisoning, and other known AI-specific attack vectors.

• Accuracy thresholds are not fixed by the regulation -- they are context-dependent and set by providers
• Article 15(2) requires accuracy metrics to be declared in technical documentation and the instructions for use
• Model monitoring for performance degradation over time is part of post-market monitoring under Article 72
• Adversarial robustness requirements align with the NIS2 Directive cybersecurity obligations for critical entities
• Backup technical solutions must be provided for systems where accuracy failures could cause serious harm