Accuracy, Robustness and Cybersecurity for High-Risk AI Systems

This article requires providers of high-risk AI systems to design and develop their systems to achieve appropriate levels of accuracy, robustness, and cybersecurity, ensuring consistent performance throughout the system's lifecycle. Providers must declare the accuracy levels and relevant metrics in the system's instructions for use. Systems should be resilient to errors and faults, with backup measures in place, and designed to minimize the risk of biased outputs. Additionally, providers must ensure the systems are secure against unauthorized exploitation of vulnerabilities.