TLDR: The European Commission has officially recognized a voluntary Code of Practice as a compliance tool for the AI Act’s rules on general-purpose AI models, effective August 2, 2025. A list of initial signatories, including major AI companies like OpenAI, Google, and Microsoft, has been published, while Meta has opted not to sign.
Brussels, Belgium – August 1, 2025 – The European Commission today announced a significant step in the regulation of artificial intelligence, formally recognizing a voluntary Code of Practice as an official compliance tool for the upcoming AI Act’s rules on general-purpose AI (GPAI) models. These obligations are set to take effect across the European Union tomorrow, August 2, 2025, ushering in new requirements for transparency, safety, and accountability in AI systems.
The GPAI Code of Practice, developed by independent experts, is designed to assist providers of general-purpose AI models in demonstrating compliance with the AI Act. Providers who sign and adhere to this Code are expected to benefit from reduced regulatory burden and increased legal certainty. Conversely, those who do not sign may face increased regulatory inquiries and be required to provide detailed gap analyses comparing their compliance frameworks with the Code’s measures.
The Commission has also released a list of initial signatories to the Code, which includes prominent AI developers such as OpenAI, Anthropic, Mistral, Microsoft, Google, IBM, and Amazon. Elon Musk’s xAI has also signed, though specifically only the section pertaining to safety. These companies have committed to various stipulations, including drawing up and implementing copyright policies compliant with EU law, continuously assessing and mitigating systematic risks associated with their AI models, and reporting serious incidents to the EU AI Office.
However, not all major players have joined. Meta’s Chief Global Affairs Officer, Joel Kaplan, announced on July 18, 2025, that Meta would decline to sign the Code of Practice, citing concerns that it overreaches the scope of the AI Act. This decision highlights a potential divergence in industry approaches to AI regulation.
The AI Act’s obligations for GPAI models define them as those trained with over 10^23 FLOP and capable of generating language, displaying significant generality, and performing a wide range of distinct tasks. The Code of Practice addresses key areas such as copyright and transparency, with advanced models deemed to have ‘systemic risk’ facing additional voluntary obligations concerning safety and security. Signatories are required to respect data access restrictions for training models, implement technical safeguards against copyright infringement, and publicly disclose information on new AI models at launch, as well as provide data to the EU AI Office and national authorities upon request.
Also Read:
- Microsoft Endorses EU’s Generative AI Code of Practice Amidst Industry Divide
- xAI Commits to EU’s Generative AI Safety Code
To further assist providers, the Commission has published guidelines clarifying who must comply with these obligations and a template to help summarize the data used to train their models. The widespread adoption of the GPAI Code of Practice remains to be seen, but the Commission strongly encourages its unconditional and full uptake as a ‘straightforward way of demonstrating compliance’ that will streamline enforcement activities and foster trust.
