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

On 14 November 2024, the European Commission published the first draft of the General-Purpose Artificial Intelligence (AI) Code of Practice (Code), including testing requirements. The Code is aligned with Articles 53(1)(a) and 55(1)(a) of the Artificial Intelligence Act, setting specific obligations for providers of general-purpose AI models regarding testing and evaluation. Signatories are required to document testing procedures and results as part of their technical documentation and Safety and Security Reports (SSR). Providers of general-purpose AI models must document their testing procedures and results within their technical documentation and Safety and Security Reports (SSR). For models classified as posing systemic risk, testing obligations include adversarial evaluations, capability elicitation through fine-tuning and scaffolding techniques, and generalisation assessments across diverse usage scenarios. All evaluations must adhere to stringent scientific standards, include independent expert review when high-severity risks are identified, and be performed continuously throughout the AI model’s development and deployment lifecycle. Testing results are required to guide deployment decisions and be submitted transparently to the European AI Office in standardised formats, explicitly detailing uncertainty measures and methodological limitations.