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 cutting-edge AI models, according to AI Minister Kanishka Narayan. Narayan made the announcement on Monday, revealing that the institute is collaborating with the US government and prominent developers, such as Google, to evaluate the safety of these advanced systems.

Britain, which lacks any major AI developers, aims to assume a leading role in assessing and managing the risks associated with increasingly powerful AI systems. Concerns about AI safety have surged after developers like Anthropic and OpenAI revealed instances of rogue or undesired behavior from AI agents. Some industry experts have cautioned that the technology could pose an existential threat to humanity.

The institute's assessment of the latest models, including Gemini 4 Argon, follows a report last month that the White House had asked OpenAI and Anthropic to withhold new models from British testers until a US review was completed. Narayan confirmed that AI scientists are performing thorough evaluations of the newest generation of models to pinpoint risks early, test robust safeguards, and contribute to establishing international safety standards under voluntary agreements with major AI developers.

The institute gains access to models before their deployment, enabling it to assess their capabilities and risks and contribute to emerging global safety guidelines. Microsoft President Brad Smith, in a recent BBC interview, noted that Britain is well-positioned to lead work on AI safety guidelines. In a September letter, the AI Security Institute stated that no organization outside the US had access to Anthropic's Mythos 5.1 model.

When questioned about access to frontier AI systems, a matter the ministers have not confirmed the institute had lost, Narayan emphasized the government's long-standing relationship with the US, built on trust. Narayan stated that while relying on others' technology, one must be clear-eyed about the permanence of any outcome, but he remains confident that, due to the conversations held, the institute will still be able to test the latest generation of models.