Britain's AI safety testers are assessing the latest models again, minister confirms
Kanishka Narayan said scientists at the AI Security Institute are conducting rigorous evaluations of the newest generation of models, including Gemini 4 Argon, weeks after reports that Washington had asked two US developers to hold new systems back from British testers.
Britain's AI Security Institute has been given access to the latest advanced models and is working with the US government and major developers, including Google, to assess their safety, the AI minister, Kanishka Narayan, said on Monday.
Narayan confirmed that scientists at the institute are now conducting rigorous evaluations of the latest generation of models, including one named Gemini 4 Argon, in order to identify risks early and test safeguards. The announcement did not specify every model the institute is evaluating.
Under voluntary agreements with major developers, the institute receives access to models before deployment, which allows it to assess their capabilities and risks and to contribute to emerging international safety standards. Separately, in remarks reported by Computer Weekly, Narayan argued, speaking at Imperial College London's science and technology incubator, Base One, for a whole-society approach to resilience in the face of emerging AI risks.
We know how to do this for other products.Brad Smith, president of Microsoft, in a BBC interview reported by Reuters, 12 October 2026
The confirmation matters because of what preceded it. Monday's announcement came after reports last month that the White House had asked two leading US developers to withhold new models from British testers pending a US review. In a letter published in September, the institute said one major developer had made clear that no organisation outside the United States had access to its system known as Mythos 5.1.
Ministers have never confirmed that the institute lost access on a wider basis. Asked about access to frontier systems, Narayan said the Government had a long-standing relationship with the US built on trust, and he presented the current evaluations as that relationship working as intended.
The wider argument about why such testing matters was made the same day by Microsoft president Brad Smith. In a BBC interview published on Monday, Smith said AI currently faces fewer safety requirements than milk, and called for global safety rules requiring companies to inform anyone whose system is breached by a rogue AI agent, to file incident reports, and to have an emergency brake available.
Smith said the one country well placed to lead on such rules was Britain, because of the AI Security Institute, which was launched under the former prime minister, Rishi Sunak. Computer Weekly reported that he told the BBC's Today programme the UK had the trained technical capability to advance safety standards and rules.
Scrutiny of AI safety has intensified after developers disclosed instances of rogue or undesired behaviour by AI agents, and some industry figures have warned that the technology could pose an existential threat to humanity. Britain is not home to any major AI developer, which is precisely why the institute's access matters: its claim to a leading role in assessing the risks of powerful systems depends on developers continuing to open their doors before those systems are released.
None to date. The Journal corrects errors promptly and transparently. Read our corrections policy.
