- The EU Commission published guidelines for general-purpose AI models on July 18, 2025; compliance creates a presumption of conformity.
- The Code of Practice for GPAIM was finalized on July 10, 2025; it offers practical implementation aids, but is not legally binding.
- GPAIM providers can join the code; some companies (e.g. Meta) reject this and criticize legal uncertainties.
The AI Act provides for both general Guidelines of the Commission before (“Guidelines”; Art. 96) also Codes of Practice specifically for providers of AI models for general purposes (GPAIM; the “Codes of Conduct”; Art. 56).
In contrast to the Guidelines, the Codes of Practice are not published by the Commission, but by the Industry – under the guidance of the AI Office – and are less concerned with the interpretation of the relevant provisions of the AI Act than with possible practical implementation measures.
Guidelines for GPAIM providers
In preparation for the obligations of GPAIM providers coming into effect on August 2, 2025 (Art. 113), the EU Commission has now on July 18, 2025 the guidelines for general-purpose AI models published:
The guidelines are not legally binding; the binding interpretation of the AI Act is reserved to the ECJ. However, compliance with them gives rise to a presumption of conformity with the obligations for providers of GPAI models, and the EU Commission will adhere to the Code of Practice in its enforcement of the AI Act.
Code of Practice for GPAIM providers: final version
At the same time July 10, 2025 the Code of Practice published in its final version.
It does not establish any binding rules either, but is intended to provide the industry with assistance in complying with the obligations of GPAIM providers and demonstrating compliance. The Code of Practice consists of three chapters, Transparency and Copyright (for all providers) and Security (for GPAIM providers with systemic risks). The Member States and the Commission now have to assess the appropriateness of the Code of Practice.
GPAIM providers are invited to join the Code of Practice (Art. 56 para. 7). Meta has has already announced that it will refrain from:
“Europe is heading down the wrong path on AI,” Kaplan posted in a statement. “We have carefully reviewed the European Commission’s Code of Practice for general-purpose AI (GPAI) models and Meta won’t be signing it. This Code introduces a number of legal uncertainties for model developers, as well as measures which go far beyond the scope of the AI Act.”