UPDATE — SEPTEMBER 2025: The European AI Office has advanced from drafting to implementation of the General-Purpose AI (GPAI) Code of Practice, a central tool for guiding providers of large-scale AI models. The first draft, published in September 2024, went through stakeholder discussions before the finalized Code was released in April 2025. It was formally endorsed by the European Commission in May.
The Code establishes standards for transparency, systemic risk management, copyright practices, and scaled compliance for small- and medium-sized enterprises. It introduces standardized disclosure templates, such as model cards aligned with ISO/IEC work. It also introduces proportionate obligations to ensure startups can comply without prohibitive costs. For open-source AI, the Code exempts certain obligations but still expects disclosures where systemic-risk thresholds are met.
By July 2025, the Commission had begun monitoring how major model providers were applying the Code’s provisions. This process treats the Code as a “bridging instrument” until the EU AI Act’s GPAI obligations. It becomes legally binding in August 2026. Providers are being urged to adopt risk assessment and stress-testing protocols voluntarily ahead of enforcement.
The AI Office is also testing the Code’s practicality with compliance workshops targeted at SMEs, start-ups, and universities. These sessions aim to validate whether proportionate obligations and exemptions work as intended. Meanwhile, international alignment has been prioritized: the Office is incorporating inputs from the G7 Hiroshima AI Process and OECD AI Principles into planning for a second iteration of the Code.
Key substantive changes since the draft include stronger systemic risk controls for models trained above 10^25 FLOPs, clarified copyright provenance statements, and expanded technical benchmarks for model evaluation and red-teaming.
ORIGINAL NEWS STORY:
EU Releases First Draft of General-Purpose AI Code of Practice, Inviting Stakeholder Feedback on AI Governance Standards
The European AI Office has released the first draft of the General-Purpose AI Code of Practice, marking an important step in the EU’s effort to set clear expectations for developers of large-scale AI models. Drafted by independent experts, the document will now move through a months-long review involving more than 1,000 stakeholders. A final version is expected in April 2025.
How the Draft Was Created
The initial text was shaped by four working groups focused on transparency, technical safeguards, copyright, and systemic risk. Chairs and Vice-Chairs from these groups contributed guidance based on both European and international best practices. Their work forms the foundation for what will become a practical framework to steer GPAI providers toward responsible development and deployment. The draft outlines a phased process for refinement. Four iterative rounds of edits will help the AI Office incorporate lessons from workshops, written submissions, and expert briefings. To support this process, the current version includes guiding principles, performance metrics, and open questions that invite targeted feedback from participants.
Early Priorities in the Draft Code
The draft highlights several early objectives. It calls for stronger transparency in AI operations, clearer copyright standards, and a structured taxonomy of risks for advanced GPAI systems. These risk categories will guide future requirements around evaluation, stress testing, and documentation. The Code also proposes proportional compliance measures for small- and medium-sized enterprises, ensuring that expectations scale with a provider’s size and business model. In addition, the draft signals flexibility for open-source AI. It suggests exemptions for some obligations, while still noting that developers must address systemic risks when their models reach a certain scale.
Stakeholder Engagement and Review Timeline
The AI Office has launched a broad engagement process. Working groups will meet separately with representatives from EU Member States, civil society, universities, and industry. Each meeting will give participants a chance to comment on recent updates and raise concerns. On November 22, the Chairs will deliver a summary of these discussions during a plenary session. To encourage wider participation, the Office has also posted the draft on its Futurium platform, where stakeholders can submit written feedback until November 28. This dual approach supports inclusive input and ensures the Code reflects real-world challenges.
Additional Resources for Providers
The AI Office has published a detailed Q&A alongside the draft. This resource explains how the Code fits within the broader EU AI Act, which took effect in August 2024. It also helps clarify how upcoming GPAI obligations may apply once the Act’s phased implementation reaches later stages.
A Key Pillar of Future EU AI Governance
Once finalized, the GPAI Code of Practice will serve as a central element in the EU’s approach to responsible AI oversight. It will establish standards for transparency, risk management, technical safeguards, and governance practices for high-impact AI models. The Code is designed to evolve over time, adapting as AI capabilities grow and as risks change.
Need Help?
If you have questions or concerns about any global guidelines, regulations and laws, don’t hesitate to reach out to BABL AI. Their Audit Experts can offer valuable insight, and ensure you’re informed and compliant.
