A leaked final text of the EU AI Act points to a notable carve-out for open-source AI in Europe. According to the source, Euractiv journalist Luca Bertuzzi has leaked the final text, and the final yes/no vote is expected to take place in the Committee of Permanent Representatives (COREPER) in early February.
The most important signal for developers is that certain third parties could avoid some compliance requirements when they provide AI tools, services, processes or components under a free and open license. The text also encourages common documentation practices, including model maps and datasheets, as a way to improve information-sharing across the AI value chain.
What the leaked text says about open-source AI
The reported language is important because it recognizes a specific role for free and open-source licence models in AI development. The source says software and data, including models, can contribute to research and innovation when released under a license that allows open sharing and broad user freedoms.
Those freedoms are described in practical terms: users can access, use, modify and redistribute the software, data or modified versions. That definition matters because it separates open-source AI from systems that may be available to view or use but do not grant the same level of practical control.
For the European Union, the leaked text frames this openness as more than a developer preference. It says such software and data can provide significant growth opportunities for the Union economy. In other words, the draft does not treat open-source AI only as a technical distribution choice; it also connects it to research, innovation and market development.
Where the compliance relief appears to apply
The reported exemption is aimed at third parties that provide AI tools, services, processes or components under a free and open license. The source says those third parties will be exempt from certain compliance requirements.
That wording is narrow in two ways. First, it applies to certain compliance requirements, not every obligation. Second, it is tied to the way the AI tools, services, processes or components are licensed and shared.
The source also states that foundational models are exceptions. That means the open-source signal should not be read as a blanket exemption for every AI model or every AI developer. The leaked text appears to distinguish between open components in the AI value chain and foundational models, which are treated differently in the reported summary.
For teams building with open-source AI in Europe, the practical takeaway is cautious but meaningful. The leaked text suggests that open licensing can reduce some compliance exposure for certain third-party providers, while still leaving important boundaries in place.
Why documentation still matters
The leaked EU AI Act text does not simply point to exemptions. It also encourages developers to use common documentation practices such as model maps and datasheets.
That is a significant companion idea. Open licensing can make AI tools and components easier to share, inspect and adapt. Documentation can make them easier to understand, evaluate and integrate responsibly.
The source says these practices should facilitate the exchange of information in the AI value chain. In plain terms, when developers document what they build, downstream users have a better chance of understanding how an AI component fits into a larger system.
The reported goal is also tied to trust. The source says the documentation practices should promote the development of trustworthy AI systems in the European Union. That connects open-source AI with a broader policy objective: making the movement of information through AI development clearer and more useful.
What developers and AI teams should watch
The next formal step named in the source is the expected yes/no vote in COREPER in early February. Until that vote happens, the leaked final text remains a strong signal rather than a completed outcome.
Still, the direction described in the source is clear enough to matter. The reported text gives open-source AI a recognized place in the EU AI Act discussion and suggests that free and open licensing may reduce certain compliance burdens for third-party providers.
For AI teams, the most relevant points are:
- Open license terms matter. The reported exemption depends on tools, services, processes or components being provided under a free and open license.
- User freedoms matter. The source describes open-source software and data as openly shared and freely accessible, usable, modifiable and redistributable.
- Foundational models are treated separately. The source says foundational models are exceptions to the exemption.
- Documentation remains important. Model maps and datasheets are encouraged to improve information exchange and support trustworthy AI systems.
The leaked text therefore offers a balanced signal. It appears to acknowledge that open-source AI can support research, innovation and economic opportunity, while still expecting better documentation and preserving a separate treatment for foundational models.
For Europe’s AI ecosystem, that combination could shape how open-source components are shared and used. The final impact will depend on the outcome of the COREPER vote and how the final EU AI Act text is applied after that step.