Share This Article
Negotiations on the EU AI Act progress with new rules applicable to Foundation Models and General Purpose AI, and more stringent obligations for copyright protection.
Spain is spearheading the EU’s AI Act negotiations. As the legislative process nears its final ‘trilogue’ phase, Spain’s proposed obligations focus on transparency and compliance with EU law for ‘foundation models’ and ‘General Purpose AI.’
Foundation models, like GPT-4, will need clear technical documentation and respect for copyright laws, ensuring training data is used with creators’ consent, if rightsholders have opted out of the copyright exception for text and data mining. Providers must also publish summaries of their training content, maintaining transparency about copyright management ➡️ This is the first case when copyright issues are tackled so much in detail and if the obligation remains unchanged, it will be problematic for major providers
High impact foundation models, which are more advanced and pose systemic risks, the EU will set stricter controls. Providers may have to undergo ‘red-teaming’ to identify weaknesses, potentially leading to independent audits mandated by the AI Office. High-impact models will also need annual risk assessments to evaluate potential negative impacts, including those on democracy. The Commission will keep regulations updated with technological advances ➡️ It shall be assessed whether such evaluations shall be done by an independent third party that would be burdensome
General Purpose AI systems, which can be used in various ways like ChatGPT, come with their own set of rules. Providers must clarify high-risk uses, provide detailed documentation for licensing agreements, and take steps to prevent misuse ➡️ The clear definition of the scope of this category will be crucial.
It will be an intense month of negotiations, and I will try to keep you posted. On a similar topic, you can read the article “Exciting Developments on the EU AI Act“.
