EU AI Act Faces Collapse Amid Foundation Model Disputes
Germany, France, and Italy push for self-regulation of foundation models, threatening to derail the EU AI Act as crucial trilogue negotiations run out of time.
Ongoing trilogue negotiations over the EU AI Act face a major crisis as Germany, France, and Italy reject comprehensive regulation for foundation models. Instead of adhering to a previously agreed graduated approach that applies stricter rules to more powerful AI systems, these countries push for self-regulation through a code of conduct to protect their domestic AI companies. This sudden reversal threatens to undermine the entire legislative effort just as the negotiation cycle nears its planned conclusion.
In response to this pushback, the European Commission proposes a compromise text that maintains the tiered regulatory structure but significantly weakens the overall rules. This watered-down proposal now serves as the foundation for further discussions, though it faces expected resistance from the European Parliament, which generally advocates for stronger AI oversight. The disagreement centers specifically on how to handle general-purpose AI systems like ChatGPT, which train on massive amounts of data to perform a wide variety of complex tasks.
Time is running critically short for lawmakers to reach a consensus before the final trilogue in December. Failure to secure an agreement before Belgium assumes the Council Presidency and the European Parliament elections take place in summer 2024 results in severe delays or the complete collapse of the legislation. As the EU strives to maintain its position as a global pioneer in AI regulation, this internal division puts significant pressure on negotiators to finalize a workable solution.