UK AI Security Institute again testing frontier models after US access freeze ends

Britain’s AI Security Institute (AISI) is again conducting safety evaluations of the latest frontier AI models and working with the US government and major developers, AI minister Kanishka Narayan said on 12 October 2026, Reuters reported.

What changed

Under voluntary agreements, AISI has received pre-deployment access so it can assess capabilities and risks. Last month, reports said the White House asked OpenAI and Anthropic to withhold new models from British testers pending a US review. In a September letter, AISI said Anthropic had made clear that no organisation outside the US had access to Claude Mythos 5.1. The Telegraph reported that the freeze was reversed after lobbying by Narayan, restoring pre-release access involving Anthropic, Google DeepMind and OpenAI.

What Narayan confirmed

Narayan said AISI scientists are now rigorously evaluating the latest generation of models, including Gemini 4 Argon, to identify risks early and test safeguards. He did not list every model under review. Asked about permanence of access, he said Britain must be “clear-eyed” when reliant on someone else’s technology, but was confident conversations with the US would keep testing possible.

Context

The arrangement remains voluntary, not a binding treaty. Microsoft president Brad Smith told the BBC that Britain is well placed to help lead AI safety guidelines. The episode shows how quickly pre-release access can be paused or restored for geopolitical reasons.