by Alimat Aliyeva
The United States has restored the United Kingdom’s access to advanced artificial intelligence models developed by OpenAI and Anthropic, allowing British specialists to resume safety assessments of some of the world’s most powerful AI systems.
According to media reports, the decision followed a reassessment of Washington’s position on sharing advanced AI technology with British authorities.
In late September, the administration of US President Donald Trump reportedly instructed OpenAI and Anthropic not to provide the latest versions of their models to the UK’s AI Safety Institute (AISI) while US officials reviewed the arrangement. The restriction raised concerns about the ability of British experts to independently evaluate emerging AI technologies before they became publicly available.
Under the revised decision, the AISI may once again receive access to advanced models from American technology companies ahead of their official release. This would allow researchers to examine potential vulnerabilities, assess harmful capabilities, and identify safety risks before the systems are made available to a wider audience.
Reports suggest that pressure from Kanishka Narayan, the UK’s Minister for AI, played a role in Washington’s decision to reverse course. Access to unreleased models is particularly valuable because safety researchers can identify weaknesses before a system reaches millions of users.
The agreement highlights the growing importance of international cooperation in AI safety. As AI systems become more capable, governments are increasingly seeking ways to evaluate their risks before deployment. Early access allows independent specialists to test how models respond to dangerous requests, whether they can be manipulated into bypassing safeguards, and how reliably they follow safety rules.
The development also underscores the close technological relationship between the United States and the United Kingdom. Both countries have been working to strengthen AI safety research while balancing national security concerns with the need for scientific collaboration.
One of the key challenges is that AI models are evolving rapidly, making traditional testing methods increasingly difficult to keep up with. By allowing researchers to examine systems before their public release, governments can potentially identify serious problems earlier and give developers more time to address them.
However, access alone does not guarantee that an AI system is safe. The effectiveness of these assessments will depend on the depth of testing, the independence of researchers, and whether developers act on the findings before launching their products.