European Union: European Commission published second draft of the General-Purpose AI Code of Practice, including testing requirement

On 19 December 2024, the European Commission published the second draft of the General-Purpose Artificial Intelligence (AI) Code of Practice (Code), which establishes testing requirements. The Code aligns with Article 53(1)(a) and Article 55(1)(a) of the AI Act, requiring providers of general-purpose AI models to conduct and document thorough testing for safety and reliability. Providers are required to comprehensively document the entire testing process and outcomes, including adversarial testing, to assess the model’s behaviour, reliability, and potential systemic risks. Testing must be designed to reflect the model’s intended capabilities, usage contexts, and exposure to high-risk tasks, with all procedures recorded in the technical documentation. For models presenting systemic risks, testing must incorporate standardised protocols and tools aligned with state-of-the-art safety evaluation practices.