European Union: Published second draft of the General-Purpose AI Code of Practice, including quality of service requirement

On 19 December 2024, the European AI Office published the second draft of the General-Purpose Artificial Intelligence (AI) Code of Practice (the Code), which addresses Quality of Service requirements through the inclusion of detailed Commitments and Measures aimed at ensuring model reliability, robustness, and lawful operation throughout the AI value chain. While the term "Quality of Service" is not expressly used, the Code incorporates analogous requirements within documentation and lifecycle obligations, particularly for general-purpose AI models with systemic risk. Commitment 6 on risk assessment and mitigation along the model lifecycle and Commitments 11 to 13 on technical mitigations mandate the adoption of effective safeguards to ensure the intended and reliable performance of the model. These include documenting design specifications, assessing deployment contexts, monitoring system behaviour, and evaluating robustness against systemic risk sources such as model misalignment, hallucination, or lack of reliability. Performance indicators must also reflect operational efficacy and adaptability to real-world environments, subject to continuous review by the AI Office. Written feedback will be received until 15 January 2025, and several discussions are planned prior to the release of the third draft.