Official name: General Purpose AI Code of Practice, Transparency Chapter
Requires general-purpose AI model providers to comply with AI Act obligations, particularly Articles 53 and 55. Obligates providers to maintain, disclose, and update model documentation for compliance assessment. Tasks the AI Office with overseeing adherence to these obligations.
Analysis summaries, actor details, and coverage mappings were LLM-classified and may contain errors.
This is a voluntary Code of Practice that uses predominantly voluntary language ('commit to', 'will', 'are encouraged') and relies on signatories choosing to adhere to it. While it references binding obligations from the AI Act, the Code itself serves as guidance for demonstrating compliance rather than creating new binding legal obligations.
The document has minimal risk domain coverage, focusing primarily on governance and transparency mechanisms rather than specific AI risks. It addresses governance failure (6.5) through documentation and oversight requirements, and touches on lack of transparency (7.4) through disclosure obligations. The document is procedural in nature, establishing compliance frameworks rather than directly addressing most risk categories.
This document does not govern specific economic sectors. Rather, it establishes horizontal transparency and documentation requirements for providers of general-purpose AI models across all sectors. The obligations apply to AI model providers regardless of the downstream sectors in which their models may be deployed.
The document primarily covers the Deploy and Operate and Monitor stages of the AI lifecycle, focusing on documentation requirements when placing models on the market and ongoing obligations to update and provide information. It implicitly touches on Build and Use Model through documentation of model characteristics, but does not substantially address earlier stages like planning, design, or data collection.
The document explicitly focuses on general-purpose AI models as its primary scope, with specific provisions for models with systemic risk. It does not explicitly mention foundation models, generative AI, predictive AI, task-specific AI, or compute thresholds. Open-source models are mentioned with an exception clause.
European Union; AI Office
The document is a Code of Practice developed under the EU AI Act framework, with the AI Office playing a central role in its development and implementation as indicated by references to the AI Office's oversight responsibilities.
AI Office; national competent authorities
The AI Office is designated as the primary enforcement body with authority to request information and assess compliance. National competent authorities also have supervisory roles, particularly for high-risk AI systems.
AI Office; national competent authorities
The AI Office monitors compliance through information requests and assessment procedures. National competent authorities also conduct supervisory monitoring, particularly for high-risk AI systems built on general-purpose AI models.
providers of general-purpose AI models; Signatories
The Code explicitly targets providers of general-purpose AI models who become Signatories to the Code. These are the entities that must comply with the documentation and transparency obligations outlined.
5 subdomains (2 Good, 3 Minimal)