GitHub and Hugging Face join forces to advocate open source in EU’s AI rules

In an open letter, tech giants GitHub and Hugging Face, along with other prominent firms like Creative Commons, are urging the European Union to reconsider its approach to upcoming regulations governing open-source artificial intelligence (AI) models.

The letter calls for a thorough review of the European AI Act, expressing concerns that certain provisions within the legislation could impede the progress of open source AI development. The tech companies argue that applying regulations designed for commercial AI products or deployed systems to open source projects would be incompatible with the collaborative nature of open source development and detrimental to individual developers and non-profit research organizations. GitHub emphasized this point in a blog post, stating that such a move would run counter to the ethos of open source practices.

The letter proposes five key suggestions to ensure that the AI Act accommodates open-source models effectively. These include clear definitions of AI components, exemptions for collaborative development on open source models from bill requirements, permissions for limited real-world testing by researchers, and proportionate regulations for “foundation models.”

Open source software allows anyone to access, modify, and improve the source code, making it a valuable asset in the realm of AI for training and deploying models.

The European Parliament passed the AI Act with a significant majority in June, but it awaits approval from the EU Council, representing 27 member states, before becoming law. Negotiations with individual EU members will follow to refine the legislation’s finer details.

The open letter recognizes the Act’s intention to regulate AI responsibly and foster innovation while addressing its risks. It encourages transparency and collaboration among various stakeholders, aiming to establish standards, oversight, and liability measures to mitigate potential harms stemming from AI technologies.