EU lawmakers and governments have agreed provisional terms for regulating artificial intelligence (AI) systems like ChatGPT, taking a step closer to clinching landmark rules governing the technology as talks continued into a second day.
As negotiations approached the 24-hour mark, all sides agreed to a provisional deal on how to regulate fast-growing generative AI systems such as ChatGPT in the early hours of Thursday, overcoming one of the biggest stumbling blocks to a final agreement, a source familiar with the matter told Reuters.
According to a document circulated among lawmakers, seen by Reuters, the European Commission would maintain a list of AI models deemed to pose a “systemic risk”, while providers of general-purpose AIs would have to publish detailed summaries of the content used to train them.
The law may also exempt free and open-source AI licences from regulation in most cases, unless for example they were deemed high-risk or being used for already banned purposes.
Terms for the other obstacle, the use of AI in biometric surveillance, and source code access were yet to be hashed out, two other sources familiar with the matter said.
The sources declined to be identified because the talks are confidential.
Earlier on Thursday, the Council of the European Union postponed a morning press conference until further notice as negotiations continue.
Amid tense debates, and frustration over a broken drinks machine, one of the sources said delegates ran out of food and coffee at about 0200 GMT.
“New day, same trilogue!”, EU industry chief Thierry Breton said in a post on social media platform X, referring to the process of negotiations between Parliament, the Commission and the Council.
In a photo with the post, EU lawmakers Dragos Tudorache and Brando Benefei – also lead AI negotiators for Parliament – are seen huddling in intense discussion with fellow lawmaker Kim van Sparrentak, who has also worked closely on the AI draft rules.
EU countries and lawmakers have been trying to finalise details of the draft rules proposed by the Commission two years ago but have struggled to keep up with the rapidly evolving technology. That made a consensus hard to achieve.
Much is riding on the new law for the bloc.
It could become the blueprint for other governments as countries seek to craft rules for their own AI industry, providing an alternative to the US’ light-touch approach and China’s interim rules.
EU countries and lawmakers are racing to get a final deal ready for a vote in spring, ahead of Parliamentary elections in June when the legislative process will grind to a halt.
Failure to do so could lead to the law being delayed and the 27-member bloc losing its first-mover advantage in regulating the technology.
Even so, it could be close to two years before any legislation comes into effect.
Years in the making
The first broad framework of the law was proposed in early 2021, almost two years before the launch of Microsoft-backed OpenAI’s ChatGPT transformed the potential uses of the technology.
OpenAI’s founder Sam Altman and computer scientists have also raised the alarm about the danger of creating powerful, high intelligent machines which could threaten humanity.
Agreeing provisional terms on foundation models – the generative AI such as OpenAI which trains on large sets of data to perform various tasks – would be a big step.
The details of what was agreed were not clear. A fourth source said there were still aspects to be thrashed out.
But a late proposal by France, Germany and Italy that makers of generative AI models should self-regulate had added a point of discord. Such a move however would benefit France-based AI company Mistral and Germany’s Aleph Alpha.
On biometric surveillance, EU lawmakers want to ban the use of AI, but governments have pushed for an exception for national security, defence and military purposes.