Hugging Face has thrown its weight behind a coordinated push to keep open source machine learning from getting tangled in the EU AI Act’s regulatory net. The company, alongside Creative Commons, Eleuther AI, GitHub, LAION, and Open Future, published a position paper arguing that the Act’s current trajectory could inadvertently punish the very development model that makes AI transparent and accessible.
The coalition isn’t asking for a free pass. Instead, they’re offering five concrete recommendations aimed at the final negotiation phase. The most pointed: clarify that developers who collaboratively build open source AI components and publish them in public repositories shouldn’t be treated the same as commercial deployers. They want the Parliament’s Recitals 12a-c and Article 2(5e) strengthened so that contribution to open repositories doesn’t trigger the Act’s heavier compliance requirements.
There’s also a push to make the research and development exception actually usable. The group argues that limiting testing in real-world conditions would cripple open development, where iterative, public experimentation is the norm. They want a hybrid approach that borrows from the Council’s text while amending the Parliament’s Article 2(5d). For foundation models specifically, they’re urging proportional requirements that distinguish between development modalities — open source approaches shouldn’t face the same obligations as closed, commercial systems.
Hugging Face’s involvement carries weight. The company has built its entire platform on community contributions, hosting hundreds of thousands of public models and datasets. If the EU gets the open source carve-out wrong, the chilling effect would be immediate: researchers and small developers could stop sharing work publicly out of fear of compliance burdens they can’t afford. That would leave the field dominated by large companies with the legal resources to navigate regulation — precisely the opposite of what the Act’s transparency goals aim to achieve.
