A technical meeting on EU regulation of artificial intelligence was adjourned on Friday (November 10th) after major EU countries called for the withdrawal of a proposed approach for foundation models. Unless the deadlock is broken in the coming days, the entire bill is in jeopardy.
The Artificial Intelligence Act is a groundbreaking piece of legislation to regulate artificial intelligence using a risk-based approach. The dossier is currently in the final stages of the legislative process, and the main EU institutions have gathered in so-called trilogues to agree on the final provisions of the law.
Foundation models have become a sticking point at this late stage of the negotiations. With the rise of ChatGPT, a popular chatbot based on OpenAI’s powerful GPT-4 model, EU policymakers have been wondering how best to cover this type of artificial intelligence in the upcoming law.
At the last political trilogue on October 24, there it seemed to be the consensus introduce rules for core models following a phased approach, namely introducing stricter rules for the most powerful ones that will have a greater impact on society.
This approach, which is similar to the Digital Markets Act (DMA) and Digital Services Act (DSA), was seen as a concession by the European Parliament, which would prefer horizontal rules for all foundation models.
The point of the multi-layered approach was to place the strictest obligations on the leading providers who are currently non-European companies. However, this approach met with increasing opposition from major European countries.
On Sunday, the Spanish presidency sent into circulation the first draft that it is putting into circulation multi-layered approach black and white for internal feedback. Co-rapporteurs of the European Parliament he replied with some modifications on Wednesday, maintaining the overall reserve structure.
However, at Thursday’s meeting of the Telecommunications Working Group, a technical body of the EU’s Council of Ministers, representatives of several member states, most notably France, Germany and Italy, opposed any regulation of the core models.
Leading the charge against any regulation of the underlying models in the AI rulebook is Mistral, a French start-up that he threw down the gauntlet to Big Tech. Cedric O, France’s former secretary of state for digital, is pushing Mistral’s lobbying efforts, arguing that The AI Act Could Kill the Company.
Meanwhile, Germany is under pressure from its own leading artificial intelligence company Aleph Alpha, which Euractiv understands has very high-level connections with the German establishment. All these companies fear that EU regulations could put them on the back burner compared to their American and Chinese competitors.
Despite the efforts of the Spanish Presidency to mediate an agreement with the European Parliament, in the face of these strong positions of the political powers, the Spaniards proposed a general review of the provisions on the foundation models.
Pressed for an hour and a half about the reasons for such a change of direction, arguments made included that this multi-layered approach would amount to “regulation within a regulation” and could threaten innovation and a risk-based approach.
The European Commission originally proposed a multi-layered approach, whereby the EU executive would be at the forefront of implementing the underlying models. However, the Commission did not defend him during the technical meeting.
The representatives of the European Parliament ended the meeting two hours early because “there was nothing to discuss”. Euractiv understands that regulating the establishment model is a red line for parliamentarians without which no agreement can be reached.
“The ball is now in the court of the Council to come up with a proposal,” a parliament official told Euractiv on condition of anonymity, stressing that the presidency had no alternative solution to the multi-layered approach.
Another EU official also anonymously told Euractiv that the presidency is trying to convince reluctant member states, which are against regulation of systemic actors at the model level, but not at the system level.
At the same time, Euractiv understands that a growing faction within the most undecided member states opposes the AI Law as a whole, seeing it as over-regulated. Indeed, if a solution is not found soon, the entire law could be in jeopardy.
EU policymakers were expected to strike a political deal at the next trialogue on December 6, meaning the landing zones for the most critical parts should be more or less in sight by the end of November.
If no agreement is reached in December, the outgoing Spanish presidency would have no incentive to continue working on the technical level, and the incoming Belgian presidency would have only a few weeks to tie up the loose ends of such a complex file before the European Parliament dissolves for EU elections next June .
Moreover, a general rethinking of the underlying model approach would also require a deep overhaul of the regulatory governance architecture and responsibilities provisions along the AI value chain, for which there may simply not be enough time.
When the AI Act was proposed in April 2021, the EU had first-mover advantage in setting the world’s international standard for AI regulation. As hype about artificial intelligence grew, policymakers in the OUR, UK and China they became more and more active.
Failure to agree on an EU AI regulation within this mandate would not only result in a loss of momentum, but would also result in Brussels losing ground compared to other jurisdictions.
The telecommunications task force is scheduled to meet again next Tuesday. Another technical meeting is scheduled for the same day among EU co-legislators. Euractiv understands that negotiations have now escalated to the highest political level in order to break the deadlock.
“The artificial intelligence law is now in order,” a third EU official told Euractiv. “Now or never.”
(Edited by Nathalie Weatherald)