As 2023 comes to a close, the clock is ticking for negotiators crafting the EU's flagship artificial intelligence legislation, the AI Act. The urgency to pass the regulation was on full display Wednesday and into the early hours of Thursday morning as negotiators finally hit the pause button after a marathon 22-hour trilogue session.
For those following the legislation's path forward, 6 Dec. marked a critical day in the potential fate of the draft bill as it is likely the final trilogue meeting between the European Commission, European Parliament and Council of the European Union.
Pressure is mounting on the EU to pass the AI Act ahead of next year's parliamentary elections and reconstitution of the European Commission. If negotiators do not agree upon a common text soon, the proposed AI Act may be delayed for at least another year. The current European Commission has long sought to lead the global stage on AI regulation, similar to how the General Data Protection Regulation influenced data protection frameworks around the world after its passage.
Negotiators hit pause button
Expectations for a political agreement ran high throughout Wednesday, especially after the Council and Parliament each announced press conferences for early on the morning of 7 December. At one point Wednesday, the Council's timeline webpage even mistakenly posted but then quickly took down an entry that an agreement had been reached.
However, after the 22-hour trilogue session, European Commission for the Internal Market Thierry Breton posted early Thursday morning that, though "Lots of progress" was made, the parties would resume work the next day at 9:00 CET. One person present at the talks, but who could only speak on condition of anonymity, told Reuters, "We are exhausted. We cannot go on like that. We need to sleep so we can reassess the texts."
For the most part, with the exception of some photos posted on social media, the trilogue negotiations were held behind closed doors. According to Euractiv, some significant progress was made on foundation models but law enforcement provisions eventually held up negotiations.
Foundation models
Negotiators found common ground on foundation models, a sticking point that nearly derailed talks last month. By early Thursday, negotiators agreed on provisional terms for regulating general purpose AI systems. According to a document 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."
AI Act negotiations had been progressing on schedule until last month when representatives from Parliament walked out of negotiations with the Council over regulation of foundation models, also known as general purpose AI systems. At a technical meeting meeting 10 Nov., several large EU countries, including France and Germany, asked that the proposed approach to regulate foundation models be retracted. The issue involves home-grown AI startups, such as France's Mistral, which are concerned that the AI Act would stifle innovation in the EU and effectively "kill" EU-based AI startups.
Since then, the Spanish presidency had worked to mediate negotiations between the Parliament and the Council.
Law enforcement and national security
By the waning hours of the marathon session, there was "strong disagreement in the law enforcement chapter" of the rulebook, according to Euractiv's Luca Bertuzzi. Once the negotiations continue Friday, one issue will be the national security exemption, "since EU countries, led by France, asked for a broad exemption for any AI system used for military or defence purposes, including for external contractors."
Another significant issue is prohibited AI practices, including whether law enforcement can use remote biometric identification of individuals in public spaces. The Council wants law enforcement to be able to leverage such technology, among others, which the Parliament wants banned altogether. The Council has reportedly pressured members of Parliament to accept a package deal, "that is extremely close to the Council position."
There are other sticking points involving how the law would be governed by the AI Office, access to source code in cases related to high-risk AI systems, the potential fining regime by enforcement authorities, when the bill would go into effect, as well as other more technical issues.
Several civil society groups expressed concern that Parliament was being pressured to accept Article 5 provisions. Access Now's Daniel Leufer called the proposed document "an absolute disgrace," saying that "Parliament cannot accept this, no deal can be made based on this." And the Centre for Democracy & Technology's Europe Director Iverna McGowen called "on EU policy makes to not make a deal for a deal's sake."
Pressure mounts ahead of parliamentary elections
European Parliament faces a shakeup with elections next year. By spring 2024, parliamentary members will begin prioritizing their reelection efforts, which take place next June, and leadership in the European Commission will change, as well. Additionally, the term for the Spanish presidency of the Council of the EU expires at the end of the month. Belgium takes over in January.
If a deal on the AI Act cannot be reached in the near term, it would likely not be taken up again until after the elections, which pushes potential legislation back at least another year. As Charlie Hawes of Bristows pointed out, negotiations would restart "in the context of a Hungarian presidency, a slate of new Commissioner appointments and an EU Parliament with a potentially (very) different political composition."
Last month in Brussels, the IAPP spoke with AI Act co-rapporteur Dragoș Tudorache and Kai Zenner, Head of Office for German MEP Axel Voss, about the AI Act negotiations. Those conversations were recorded for The Privacy Advisor Podcast.
This story has been updated since it originally published 6 December.
