US and UK Refuse to Sign Paris AI Summit Agreement
At the AI Action Summit in Paris, roughly 60 countries — including China — signed a joint declaration. The United States and the United Kingdom held back, as JD Vance warned against overregulation in his first major foreign policy speech.
The AI Action Summit held over the past two days in Paris has closed with a visible rift between the world's major tech powers. Roughly 60 countries, including China, France, India and several European partners, signed a joint declaration on artificial intelligence today. The United States and the United Kingdom did not.
The summit, co-chaired by Emmanuel Macron and Indian Prime Minister Narendra Modi, is the third of its kind, following Bletchley Park in the UK in November 2023 and Seoul in May 2024. With two of the format's original architects now declining to sign the final text, the process born at Bletchley to coordinate global responses to frontier AI risks appears to have reached a turning point.
Vance's speech
The day's political headline-maker was JD Vance. The US vice president delivered his first major foreign policy speech in Paris since taking office, devoting it entirely to laying out the Trump administration's stance on international AI governance. His message was a direct warning against what he sees as regulatory overreach, aimed in particular at European frameworks: excessive regulation, he said, "would kill a transformative industry just as it's taking off."
Vance argued that Washington's priority is preserving American leadership in AI against rivals like China, and that regulatory restrictions — whether European or from any other bloc — risk slowing that race without actually reducing the risks they claim to address. The stance clashes head-on with the spirit that launched this summit cycle at Bletchley Park, which centered on the existential risks of the most advanced models and the need for international oversight mechanisms.
The UK backs out too
The bigger surprise wasn't the US position, which was predictable given the Trump administration's pivot toward tech deregulation, but that the United Kingdom — host of the first summit and de facto co-author of the entire process — also declined to sign. Keir Starmer's government said the declaration failed to adequately address national security concerns and didn't offer concrete enough global governance to warrant its signature.
It's a more nuanced position than Washington's: London isn't rejecting the framework of international cooperation itself, just the specific content of this text. But the practical effect is the same — the two countries that did the most to launch the idea of "AI Safety Summits" focused on frontier model safety are now left out of the first major multilateral agreement signed since that founding phase.
From safety to "action"
The summit's very name signals the shift. Where Bletchley Park focused on risk — pre-launch safety testing, evaluation of dangerous capabilities, cooperation between labs and governments — the Paris gathering prioritized "action": investment, infrastructure, industrial adoption and competitiveness. China, which was absent from Bletchley's strictest early commitments, signed the Paris declaration without hesitation this time, a shift that also speaks volumes about how much the geopolitical AI landscape has changed in just over a year.
The fact that the United States and the United Kingdom — the two governments that led the discourse on catastrophic AI risks back in 2023 — are now the ones opting out of an international agreement, while China signs on, shows just how far the axis of the debate has shifted from safety toward industrial competition and the geopolitical race for tech leadership.
What's left of the Bletchley process
The question the Paris summit leaves open is whether the format launched at Bletchley Park survives as a genuine coordination mechanism, or whether, from here on, each bloc — the United States, the European Union, China — will move forward with its own regulatory framework with no common umbrella. The European Union continues to enforce its AI Act, with phased-in obligations for higher-risk systems; the United States has opted for a light-touch approach under the new administration; and China combines its own regulatory framework with an open push to accelerate development of its models.
Without consensus among the three major powers on a single document, the next gathering in this summit cycle — if there is one — will have to decide whether it still calls itself a safety summit, or openly acknowledges that its purpose has already changed.
