European lawmakers are yet to agree on several issues related to new artificial intelligence rules ahead of a crucial meeting on Tuesday, leaving any deal off the table until December, according to four people familiar with the matter.
The draft AI rules have to be agreed by the European Parliament and European Union member states.
They have so far been discussed three times in trilogues, which are meetings between parliament and EU states to thrash out the final versions of laws.
A fourth trilogue meeting will be held on Tuesday, a day after EU lawmakers are scheduled to discuss their negotiating stance around foundation models and high-risk AI systems, sources said.
Foundation models, such as OpenAI’s ChatGPT, are AI systems that are trained on large sets of data, with the ability to learn from new data to perform a variety of tasks.
Spain, which holds the EU presidency until December, has been pushing for a deal and has proposed compromises in a bid to speed up the process.
These include a tiered approach for regulating foundation models, defined as those with more 45 million users, according to a draft seen by Reuters.
Spain also wants additional obligations for very capable foundation models (VCFM), such as ChatGPT, including regular vetting to uncover potential vulnerabilities.
Opponents say that smaller platforms can be equally risky.
Spain said it had consulted other EU countries on potential compromises ahead of the fourth trilogue. However, a final agreement is unlikely to be reached in that meeting, the sources said.
A fifth trilogue is due to be held in early December.
Failure to reach a deal then could push negotiations to early next year. Discussions could then be further de-railed by the European parliament elections in June.
Several lawmakers, including EU industry chief Thierry Breton and co-rapporteurs for the AI Act, Dragoș Tudorache and Brando Benifei, had expressed hopes that the draft will be approved before the end of the year.
The EU started working on the draft AI Act in 2021. In May this year, the European parliament agreed on draft legislation including new rules around the use of facial recognition, biometric surveillance, and other AI applications.
Under the proposals, AI tools will be classified according to their perceived level of risk, from low to unacceptable.
Governments and companies using these tools will have different obligations, depending on the classification.