EVIAN-LES-BAINS, France — The West should come together to rein in the most capable artificial intelligence models to keep China at bay, leaders and top AI CEOs agreed in talks on Wednesday.

As leaders of the seven richest countries spent more than two hours around a table with 11 of the richest men in the world, there was little to show that Washington had just suspended access to Anthropic’s latest models, creating tensions between the U.S. and its allies, while throwing America’s own AI policies into a tailspin.

The leaders and top AI executives displayed a spirit of collaboration, according to interviews and readouts from five people in attendance at the G7 meeting — uniting around a common enemy in China.

The signs from the room were that the Anthropic saga, rather than dividing nations, could give a new boost to global efforts to keep AI in check, which have been floundering since U.S. President Donald Trump took office for a second time. This optimism played out even as Trump’s administration had yet to announce any truce in its latest confrontation with Anthropic, one of the United States’ top AI developers.

China was mentioned multiple times during the talks, according to detailed notes shared by one participant, reflecting fears that while American companies gained an early lead on artificial intelligence, China is becoming a real competitor in part due to the energy supplies needed to build the most powerful models.

“Everybody knows that it’s really important that G7 countries and its allies band together — that we win in the AI race and .. don’t let autocratic countries take the winning position,” Victor Riparbelli, CEO of U.K.-based Synthesia, told POLITICO after the meeting.

Riparbelli was one of a group of 11 CEOs present that included Anthropic’s Dario Amodei, OpenAI’s Sam Altman and DeepMind’s Demis Hassabis, as well as Arthur Mensch, the CEO of French AI darling Mistral.

Despite very public bust-ups in recent months between Amodei and Trump over Anthropic’s highly capable models, as well as major tensions between companies, French President Emmanuel Macron managed to keep those differences under wraps and walked out of the meeting he hosted with broad consensus to collaborate further. 

“We had a great meeting on AI,” Trump said after the lunch, adding the talks between the U.S. administration and Anthropic are “going fine.” 

Trump sat next to OpenAI’s Altman, seemingly joking with Altman when the meeting started, but opposite Amodei. In his remarks to the group, Trump brought up his uncle who was an engineer at MIT, according to two people present — as well as going off topic to talk about the recent ceasefire deal with Iran.

Asked if the suspension of Anthropic’s models was addressed, Mensch said to POLITICO in a phone call after the lunch: “No, it wasn’t addressed head-on.”

“Rather, it was approached from the perspective of the interdependence of supply chains. As I pointed out, maintaining this interdependence is very important because, in fact, it will be the driving force behind stability in the emerging new global economy,” the CEO said.

The takeaway from Wednesday’s talks on artificial intelligence was the idea of defining standards that the developers of frontier AI models should adhere to. | Ludovic Maron/AFP via Getty Images

The takeaway from Wednesday’s talks on artificial intelligence was the idea of defining standards that the developers of frontier AI models, such as OpenAI and Anthropic, should adhere to. The G7 already propped up a code of conduct for these companies in 2023, then under the presidency of Japan.

“The challenge before us is a frontier model,” Macron told reporters on the ground in Evian. “Better regulation is needed to avoid it falling into the hands of an authoritarian regime.”

Macron said the G7 would “create a platform for cooperation and discussion among a couple of democracies to define standards.” A G7 ministerial is planned for September to shape that. 

The European Union was one of the driving forces behind the call for more collaboration on frontier AI models and the security threats they pose. 

The bloc steered clear of making a fuss over the U.S. move on Friday to cut off EU citizens’ access to Anthropic’s models, instead pushing for deeper collaboration.

“I believe Europe and the U.S. should work together on AI,” European Commission President Ursula von der Leyen said about the meeting, adding that both have “shared security interests and a common responsibility to lead.” Under the bloc’s AI law, AI companies already face strict obligations to evaluate and test their models and inform regulators.

Tech CEOs present in the room also promoted the idea of frontier AI standards — or walked away with the impression that it’s the most realistic outcome for now. 

Altman pushed for an international discussion forum to help establish global standards for evaluating and testing models. There was a “coalescing” among leaders and executives in the room to develop such a forum, Chris Lehane, OpenAI’s chief global affairs officer, told reporters after the meeting. 

Riparbelli, the CEO of U.K.-based Synthesia, said standards would probably have the “highest impact” and would be “the most realistic to agree on.” 

But others were more skeptical, especially on whether the U.S. would take the steering wheel in developing standards, since most of the frontier labs in the West are based in the U.S. 

“We still need to be very wary,” Mensch told POLITICO. “Because if there’s one way to establish a foothold and become unassailable in countries other than the United States, it’s by owning the standards and enforcing them.” 

“It’s true that American actors tend to make that argument, which is, ‘Trust us—we’ll set standards here, and then we’ll export them to you.’”