The EU is facing Big Tech pressure to revamp its AI law
DUBLIN — The European Union’s two-year-old AI law was once considered too far ahead of the curve. Now U.S. tech executives say it’s outdated.
Top bosses from Microsoft and Google have openly questioned whether the EU’s 2024 Artificial Intelligence Act still fits the purpose, after a summer of incidents involving AI agents going rogue and a growing chorus of warnings about the technology’s risks to humanity.
In an exclusive interview for Decoded by POLITICO on the sidelines of the international AI summit in Dublin, Microsoft President Brad Smith said European officials should reconsider the relevance of the current rules.
“One of the things that is true of every law, of every regulation over time is it’s written, it’s interpreted and then the facts change, the technology evolves, and then regulations are often reinterpreted,” Smith said.
As U.S. tech giants grow concerned about the power of their advanced AI models, they are pressing governments worldwide for new rules that would slow them down — a surprising shift from an industry that has long railed against regulation. After ridiculing Europe’s perceived obsession with rules, industry leaders now argue that the AI they are building is so scary that Europe should regulate it even more.
“If you legislate for something three years ago, it would be out of date today, and actually probably won’t even address what the real problems are,” Demis Hassabis, co-founder of Google’s DeepMind, said on stage at the event, without referring to the AI Act.
That’s sparking an open clash of ideas with top European Commission officials, who insist that the law is strong enough to handle the security risks of the newest AI models. EU tech chief Henna Virkkunen said the law is “well-equipped” to handle AI agents, even though agents weren’t mainstream when the law was written in 2023.
“We are satisfied for now; we have an appropriate system of regulation in place,” EU Justice Commissioner Michael McGrath, who oversees data protection, told reporters in a huddle on the sidelines in Dublin, responding to POLITICO’s question on whether the law is fit for purpose.
He admitted, though, that it’s vital to keep the rules “future-proof.”
Three gaps
Most talks on the AI Act happened in 2023, and the law mentions AI agents only briefly, even though they drive more than half of current AI workflows in Europe, according to OpenAI data released Tuesday.
EU legislators already agreed to a first change to the law earlier this year, delaying some of the rules and adding new bans, such as those against sexualized deepfakes. That followed criticism that the EU actually moved too quickly to regulate the nascent technology, beginning deliberations before ChatGPT’s Nov. 2022 launch.
To avoid being outpaced by technological developments, the law regulates AI use cases based on the risk they pose — avoiding the need to specify AI technologies.
But some EU governments have also concluded that the AI Act has loopholes, and that more work is needed to keep the rules up to date with the rapid pace of AI development.
Hungary told other EU countries that the AI Act is a “solid foundation” but that “three gaps nevertheless remain,” according to an October paper seen by POLITICO.
One loophole is whether the rules apply to AI models that companies use and test internally but have not yet placed on the market. “The AI Act largely applies once a model is placed on the market,” reads the paper. “The 2026 incidents occurred before that point.”
The Hungarians also worry whether authorities have enough information on when models are being trained and tested, and whether a company can be held liable if its agents do harm. “We consider that this may require legislative action and therefore propose that the necessary be taken at Union level,” it said.
Lukáš Kačena, the Czech Government’s AI envoy, said that the EU’s AI rules are a “great start” but will need updates.
“The AI Act will never be enough. Since the technology is so rapidly evolving, we will need to adapt to the current situation,” he said. “This is never going to end.”
Brendan Bordelon contributed reporting.
Read more from Brad Smith’s interview in Decoded by POLITICO, the must-read briefing on the people, power moves, policy battles and market shifts defining what’s next in global technology. Sign up to subscribe to the newsletter and listen to the podcast.
