The EU Commission announces the publication of general purpose AI code of practice
March 12, 2025 |
The European Commission has released the third draft of the General-Purpose AI Code of Practice. It includes commitments by providers of general-purpose artificial intelligence (AI) models, including:
- documentation: the signatories commit to drawing up and keeping up-to-date model documentation, including ensuring quality, security, and integrity of the documented information and providing it to providers of AI systems and to the AI Office upon request; and
- copyright policy
Providers of general-purpose AI models with systemic risk must commit to :
- adopting and implementing a Safety and Security Framework that will apply to the AI models with systemic risk, as well as detail the systemic risk assessment;
- conducting systemic risk assessment systematically at appropriate points along the entire model lifecycle;
- selecting and further characterizing systemic risks;
- determining the acceptability of the systemic risks;
- implementing technical safety mitigations along the entire model lifecycle of the model, and ensuring they are proportionate and state-of-the-art;
- mitigating systemic risks that could arise from unauthorized access to unreleased models;
- reporting to the AI Office on the safety and security of the models;
- carrying out adequacy assessments;
- implementing systemic risk responsibility allocation;
- obtaining independent external systemic risk assessments, including model evaluations;
- keeping track of, documenting, and reporting serious incidents to the AI Office and, as appropriate, to national competent authorities;
- ensuring protections on non-retaliation against any worker providing information about systemic risks;
- notifying the AI Office of relevant information and the implementation of commitments;
- carrying out documentation, as prescribed by the code of practice and the Artificial Intelligence Act (AI Act); and
- implementing public transparency on systemic risks stemming from their AI models with systemic risk.
The AI Office will:
- report on the feedback received from stakeholders on the template for an adequate public summary of the training data under Article 53(1)d) of the AI Act and outline the next steps for adopting the template; and
- publish guidance clarifying the scope of the AI Act rules for general-purpose AI, including information on:
- the definitions of general-purpose AI models;
- placement of models on the market and providers;
- exemptions for models provided under free and open-source licenses; and
- the effects of the AI Act on models placed on the market before August 2025.
The press release provides: