The US government has softened its stance on international AI cooperation, permitting American AI companies to provide early access to their models to British regulators for safety evaluation.
- US lifts ban on sharing AI models with Britain's AI Security Institute.
- Britain conducts safety testing to assess AI capabilities and misuse.
- Trump administration embraces some international AI safety collaboration.
What happened
The Trump administration reversed an earlier directive that prevented American AI developers such as OpenAI and Anthropic from sharing new AI models with the UK’s AI Security Institute before public release. This reversal permits these developers to give the British government early access to their AI systems for safety testing purposes.
The AI Security Institute in Britain, established after the landmark AI safety summit in 2023, evaluates the capabilities and potential misuse risks of cutting-edge AI technologies. By allowing earlier inspection of AI models, the UK aims to strengthen oversight and reduce harmful outcomes.
Why it matters
This policy shift signifies a more cooperative approach by the US on global AI safety and governance, after initial resistance from the Trump administration. It indicates growing recognition of the need for cross-border collaboration to address the rapid development and deployment of powerful AI systems.
The change provides US AI developers greater freedom to engage with international regulatory frameworks without fear of regulatory conflict. This collaboration can lead to more robust identification of AI risks and development of safety standards tailored to evolving technologies.
What to watch next
Observers will be watching how this closer US-UK collaboration on AI model testing influences broader international AI governance efforts, especially ahead of future multilateral discussions on AI regulation.
It will also be important to monitor whether this softer stance leads to formalized reporting requirements and security incident disclosures for AI developers, as hinted by the Trump administration’s new AI task force calling for such measures.