As part of our ongoing commitment to responsible AI development and governance, we recently participated in providing feedback on the second draft of the General-Purpose AI Code of Practice. This Code aims to implement and clarify the EU AI Act’s requirements for general-purpose AI models.
Our analysis of the second draft revealed several important considerations that warrant attention:
- The current draft includes an open-ended list of “risks for consideration.” While comprehensive, this approach might create implementation challenges similar to those we’re currently observing with the Digital Services Act, particularly regarding democratic processes and public security. There’s a delicate balance to strike between protecting against manipulation and preserving freedom of expression.
- Another significant concern relates to Measure 2.5, which addresses the exclusion of websites with potentially copyright-infringing content. The current language presents several challenges:the blanket ban approach fails to account for varying types of copyright protection, different territorial protections aren’t adequately considered.Moreover, there’s a risk of over-regulation that could limit access to legitimate public domain resources.
- We emphasize the importance of building precisely upon the foundation established by the AI Act – a regulation developed through democratic processes by elected representatives. The current draft, as we can also read in article on XYZ.pl written by Jakub Szymik, CEE Digital Democracy Watch’s CEO, delegates significant responsibility for developing precise standards to the AI Office, which could lead to uncertainty and shifting requirements. This approach might impact the clarity and effectiveness of various commitments within the Code.
As this important work continues, we remain committed to promoting broad discussion and engagement with these crucial issues.