Disputed landmark AI rules face crunch EU talks

By Foo Yun Chee, Supantha Mukherjee and Martin Coulter

BRUSSELS/STOCKHOLM/LONDON (Reuters) – European Union ambitions to take a lead in landmark rules for artificial intelligence hang in the balance as member states and lawmakers meet on Wednesday to try to hammer out a deal on biometric surveillance and how to regulate systems like ChatGPT.

If agreed, the EU’s first-of-a-kind AI Act, which was proposed by the European Commission two years ago, could serve as the benchmark for countries seeking an alternative to the United States’ light-touch approach and China’s interim rules.

Talks between EU members and lawmakers will start at 1400 GMT and are expected to run into the early hours of Thursday, with the most likely outcome a provisional deal on principles but not crucial details, five people directly involved said.

A final deal would then need to be agreed before legislation could be put in place, which could pave the way towards it becoming law before European parliamentary elections in June.

But without a deal, the AI Act is likely to be shelved due to a lack of time, resulting in the 27-member bloc losing its first-mover advantage in regulating the technology.

Alexandra van Huffelen, Dutch minister for digitalisation, told Reuters it was critical the EU finds a compromise, particularly on generative AI, by the end of the year.

“The world is watching us: citizens, stakeholders, NGOs and the private sector want us to agree on a meaningful piece of legislation regarding AI, including GPAI,” she said referring to general purpose AI systems, which have a wide range of uses.

DEMANDS

The proposed AI rules face conflicting EU demands.

The two biggest are over the use of AI in biometric surveillance and foundation models, the generative AI such as Microsoft (NASDAQ:) backed OpenAI which trains on large sets of data to perform various tasks.

EU lawmakers want to ban the use of AI in biometric surveillance, while governments want an exception for national security, defence and military purposes.

A late proposal by France, Germany and Italy to let makers of generative AI models self-regulate added more uncertainty.

EU ambassadors and lawmakers held separate preparatory meetings last week, but differences remain which could make it difficult to clinch a deal, said the people involved in the talks, who declined to be named because they are confidential.

An official from one major EU country said whatever the meeting’s outcome, there will still be a lot more work to do.