What it asks for. Article 15 requires high risk AI systems to be designed and developed to achieve an appropriate level of accuracy, robustness and cybersecurity and to perform consistently across their life cycle. Systems should be resilient to errors and to attempts by unauthorised third parties to alter their use, outputs or performance by exploiting vulnerabilities.
In practice. Providers declare accuracy metrics in instructions for use, test robustness against errors and changing inputs, handle feedback loops in systems that keep learning, and protect against AI specific attacks such as data poisoning, model poisoning, adversarial examples and model flaws. Evidence includes test reports, threat models for the AI system and security controls documentation.
Common gaps. Traditional security testing often ignores AI specific attacks. Accuracy is measured at launch but not monitored as data and conditions drift.