The European Union is nearing the possibility of becoming the first region in the world to have a comprehensive law to regulate artificial intelligence (AI). But after more than 18 hours of marathon negotiations, not all the points have yet been closed, although there is already agreement on one of the thorniest: how to regulate the foundational models on which systems such as ChatGPT are based and which, although they are considered fundamental in the evolution of technology, also generate serious doubts due to their disruptive capacity.
"Negotiations are continuing," sources close to the negotiations said. The press conferences called to present the results early Thursday morning have been postponed for the time being, in a show, however, of the will to speed up the deadlines in order to come out with an agreement for the law under the arm.
According to the sources, the fundamental obstacle for the negotiators – representatives of the States and the European Parliament, with the European Commission – to throw white smoke after a meeting that began on Wednesday at 15:00 p.m. behind closed doors is another of the main stumbling blocks in the entire process of discussing the law: the regulation of real-time biometric surveillance in public spaces through systems such as facial recognition. one of the red lines of the European Parliament, concerned about the possibility of abuses of citizens' fundamental rights by States when using these technologies.
From the negotiating mandate of the European Parliament came the decision to prohibit or restrict as much as possible the "intrusive and discriminatory uses of AI", especially biometric systems in real time or in public spaces, with very few exceptions for security reasons. The states are seeking to extend these exceptions, and there they were going to come up against a clear red line, as MEP Brando Benifei, a participant in the trialogues, warned the day before.
The position of the MEPs is much stricter than that of the states and, although the negotiations have been "difficult", there is optimism, cautious of course, about the possibility of finding a middle ground. As long as, the European Parliament stresses, the ban on predictive policing, biometric surveillance in public places and emotion recognition systems in workplaces and education systems continues to be maintained. "We need a sufficient degree of protection of fundamental rights with the necessary prohibitions on the use of [these technologies] for security and surveillance," Benifei sums up.
"Governments want a long list of exceptions to the app that we are not going to accept," the Italian said in a meeting with reporters hours before locking himself into the discussions, which he arrived at with a mandate to maintain the ban on predictive policing, biometric surveillance in public places and emotion recognition systems in workplaces and education systems.
"We need a sufficient degree of protection of fundamental rights with the necessary prohibitions when using [these technologies] for security and surveillance," according to Benifei, who said he was willing to find a "compromise" on the matter, such as in "specific cases" of police surveillance, but stressed that this required very robust "safeguards" and control of them that, in any case, they cannot be exercised by the States themselves. "We are not going to allow governments to control themselves if they respect the law, this is very, very important to us ... And we're never going to accept a deviation and not have serious control," he said.
Participants in the negotiations, such as Internal Market Commissioner Thierry Breton, sent messages on social media in the early hours of the morning showing active negotiations in a session described as an "ultramarathon". The discussions have already made it possible, according to the sources consulted, to overcome the other major obstacle to a provisional agreement on the law – which still needs to be ratified by the Council of the EU and the European Parliament before it can enter into force, at the earliest by the end of 2026 – the issue of the regulation of the foundational models. especially the more powerful ones.
More than the "what", the pulse revolved around the "how", explained sources close to the negotiation on the eve of the negotiations. Countries such as Germany, France and Italy had in recent weeks opposed the establishment of obligations by law and advocated greater self-regulation – through mandatory codes of conduct – for developers. The argument used was not to hinder innovation and competitiveness in a sector in which Europe does not want to be left behind its great rivals, the United States and China.
But MEPs, concerned about the ability of these new technologies to affect citizens' fundamental rights, had set red lines and warned of their intention to abandon the negotiations – which would have severely postponed the entire law, which is expected to be fully implemented by the end of 2026 – if sufficient safeguards were not put in place. Especially for the most powerful foundational models, those considered "systemic risk" because they have high-impact capabilities whose results may "not be known or understood at the time of their development and publication, so they may lead to systemic risks at EU level", according to the definition accepted by all parties.
The crux, Benifei also explained on the eve of the negotiations, was the way in which it is "guaranteed" for these more powerful models "that what the developers of these models do is mandatory and can be enforced". The European Parliament, he said, wanted a text clear enough to ensure that "there is no way to escape these obligations", even if they were included in a code of conduct as requested by the states, but that, in any case, "that it is not a de facto voluntary commitment, but that can be enforced".
You can follow EL PAÍS Tecnología on Facebook and X or sign up here to receive our weekly newsletter.
Subscribe to continue reading
Read without limits
I'm already a subscriber