The US and the UK sign an agreement on how to test and assess risks from emerging AI models, marking the first bilateral arrangement on AI safety in the world
Allies reach world's first bilateral deal as global governments seek to assess and regulate risks from emerging technology
Financial TimesMadhumita Murgia
Context & Ripple Effects
The agreement moves US-UK AI-safety cooperation from the broad discussions of the UK summit toward a bilateral testing relationship. It follows the summit’s tentative push for international regulatory collaboration and sits alongside an earlier industry pledge to permit pre-release national-security testing of models.
Its importance is institutional: two allied governments are defining a shared way to assess emerging-model risks, rather than relying solely on voluntary developer commitments such as the pre-release testing pledge signed by leading AI companies.
First-order effects
US and UK authorities gain a formal bilateral channel for coordinating how emerging AI-model risks are tested and assessed.
AI developers operating across the two markets face a clearer signal that government-led model evaluation is becoming part of the safety landscape, even though the reported agreement does not specify company obligations.
Second-order effects
A shared US-UK assessment approach increases pressure on other governments and regional blocs to articulate compatible testing methods rather than leave safety evaluation entirely to firms.
If similar arrangements proliferate, frontier-model governance could organize around interoperable assessment regimes among allied states, with access to models becoming a key policy lever.
That trajectory may split AI oversight into geopolitical governance blocs unless bilateral practices are reconciled through broader international frameworks.
The trend: AI governance is shifting from voluntary principles toward state-mediated access, testing, and cross-border coordination for frontier models.
The UK and US have signed a landmark agreement on AI safety. This will see the UK's @AISafetyInst join forces with the US AI Safety Institute on testing models and sharing research 🇬🇧🇺🇸 Find out more 👇
The UK-US agreement on AI safety is a significant moment for the AI Safety Institute and for the development of global safety standards on AI. Read below to find out more about how we will collaborate. (1/4)
AI safety is a major 🇺🇸-🇬🇧 priority; our partnership seeks to unleash AI's huge potential & protect our citizens. A first-of-its-kind AI safety alliance unveiled today by @SecRaimondo & @michelledonelan
This is interesting, but also worth noting that the UK is likely the very junior partner here (especially given Gina Raimondo came to the Bletchley Park Summit and announced the US was doing exactly what the UK was, just bigger, at their own party) https://www.commerce.gov/...
Building on the UK AI Safety Summit, I have just signed a landmark agreement with the US - a crucial moment in ensuring global AI safety. Both the potential benefits & risks of AI are global in scale, so this work will help us safely harness AI & improve lives across the world🌍 […
'The agreement will specifically enable the UK's new AI Safety Institute (AISI), set up in November, and its US equivalent, which is yet to begin its work, to exchange expertise through secondments of researchers from both countries. The institutes will also work together on how.…
Very proud of the landmark agreement the US and UK have signed today around joint testing of frontier AI systems. Testament to an incredible team of civil servants at the AI Safety Institute: https://www.ft.com/... [image]
This is a big deal - and further underlines the amazing progress the UK has made on state capacity in AI. Congratulations to the whole AISI team! [image]
It's definitely a good thing, but can't help but feel like the US-UK AI Safety agreement is being ever so slightly overhyped by the UK. Sharing best practices on doing capability evals is great, but “landmark”?
Today, UK Secretary of State for Science, Innovation and Technology @michelledonelan and I launched a 🇺🇸-🇬🇧 partnership on AI safety. Together, we're furthering the long-lasting special relationship between our nations and laying the groundwork to ensure we're keeping AI safe. [i…
#NEWS: U.S. and UK AI Safety Institutes to work seamlessly with each other, partnering on research, safety evaluations, and guidance for #AI safety. https://www.commerce.gov/...
This adds more meat on the bones, but isn't exactly new: It was always clear the UK and US safety institutes would work together. This is more about signalling the UK is part of the fold ahead of a wider US-EU announcement on AI collaboration later this week.
Really great to see US and UK cooperation on evals! For now, UK AISI is 10x funded compared with US AISI, but most AI labs are in the US. I'm glad these safety resources are being pooled 🇬🇧🇺🇲