According to media reports citing sources, the EU artificial intelligence (AI) bill is at risk of being shelved due to difficulties in reaching an agreement on regulating systems such as ChatGPT. After two years of negotiations, the European Parliament passed the authorization draft of the "AI Act" in June this year. The bill is currently in the tripartite negotiation stage between member states, parliaments and the European Commission, with the aim of finalizing the bill. If formally approved, the bill will become the world's first comprehensive regulation on AI.
Negotiators are scheduled to meet on Friday local time for key discussions ahead of final talks on December 6. Thierry Breton, the EU’s director of industrial policy, and Dragoș Tudorache, one of the leaders of negotiations on the AI Bill, have previously expressed the hope that the bill will be approved before the end of this year.
However, sources said that regulation of the underlying models has become a major obstacle in negotiations for the AI Bill.
Some experts and lawmakers have proposed a tiered approach to regulating the underlying model. The EU defines a base model as one with more than 45 million users. The chatbot ChatGPT is defined as a very capable base model, and such base models are subject to additional obligations, including regular reviews to detect potential vulnerabilities.
However, some lawmakers believe those smaller models may carry the same level of risk.
It is reported that the biggest challenge to reaching an agreement comes from France, Germany and Italy, which prefer to let companies developing generative AI models self-regulate rather than establish strict rules.
France persuaded Italy and Germany to back the proposal at an Oct. 30 meeting in Rome, sources said, after talks progressed smoothly and lawmakers made compromises on several other contentious areas, including regulating high-risk AI.
Breton, European lawmakers and dozens of AI researchers oppose letting companies regulate themselves. This week, researchers such as Geoffrey Hinton issued an open letter warning that self-regulation may fall far short of the standards required for the security of the underlying model.
Sources also revealed that other unresolved issues in the negotiations include the definition of AI, basic rights impact assessment, law enforcement exceptions and national security exceptions.
Since the next European Parliament elections will be held next year, if lawmakers cannot reach an agreement on the AI Bill before the end of the year, the bill may be shelved.
"If you had asked me six or seven weeks ago, I would have said we were seeing compromise on all the key issues, but now it's much more difficult," said Mark Brakel, policy director at the nonprofit Future of Life Institute.