UK given access to safety-test latest AI models, minister says

The UK's AI Security Institute has been granted access to the most recent advanced AI models, according to AI Minister Kanishka Narayan. The institute is collaborating with the US government and major tech companies such as Google to evaluate the safety of these powerful AI systems. Britain, despite not being home to any major AI developers, aims to play a leading role in assessing and managing the risks associated with increasingly sophisticated AI technology.

The focus on AI safety has grown more intense following disclosures from developers like Anthropic and OpenAI about instances of AI agents behaving in unpredictable or undesirable ways. Some industry experts have expressed concerns that this technology could potentially pose an existential threat to humanity.

This announcement, which did not detail all the models being assessed, follows reports last month that the US government has requested OpenAI and Anthropic to withhold new models from British testers until the US completes its own review. However, Narayan affirmed that AI scientists at the institute are currently conducting thorough evaluations of the latest generation of models, including Gemini 4 Argon, to detect potential risks early, test robust safeguards, and contribute to the development of international safety standards.

Under voluntary agreements with leading AI developers, the institute gains early access to models before their deployment, enabling it to assess their capabilities and risks and help shape emerging international safety standards. Microsoft President Brad Smith commented in an interview with the BBC that Britain is well-positioned to help lead work on AI safety guidelines.

In a September letter, the AI Security Institute indicated that Anthropic had made clear that no organization outside the US had access to its Mythos 5.1 model. When asked about access to frontier AI systems, a matter the ministers have never confirmed the institute has lost on a broader scale, Narayan stated that the government has a long-standing relationship with the US built on trust, but emphasized the need to remain cautious.

Despite this, he expressed high confidence that, following extensive discussions, Britain will be able to test the latest generation of AI models.