Britain's AI minister tells frontier labs “do not release the system” if they cannot control its risks, and says a breach of UK cyber defences seems likely

Kanishka Narayan, the UK's Minister for Artificial Intelligence, spoke on 12 October at Base One, Imperial College London; GOV.UK publishes the text. He addressed frontier AI developers directly: “Listen when your scientists warn of severe dangers. Investigate unexpected behaviours. And if you can't adequately control the risks, do not release the system.” He also said: “It seems likely, if not inevitable, that our own cyber defences will be breached.” And: “We will legislate wherever necessary.”

The government's release the same day sets out the package. The government will use frontier AI on its own systems, building on pilots that have “identified and addressed over 1,000 vulnerabilities”; the speech says a monitoring service has cut the time to fix weaknesses from 50 days to eight. Ministers will explore using the Cyber Security and Resilience Bill to set how serious AI incidents and near misses are identified and reported. As an example of the risk, Narayan cited an OpenAI agent in Australia earlier this year that, without being asked, found another way into a government website and reached information it was not authorised to have.

The release says AI Security Institute scientists are running “rigorous evaluations of the latest generation of advanced AI models” “following work with leading AI developers and the US Government”. The speech names Gemini 4 Argon among them and says “over the last few weeks, we've been working closely with the US government, the labs”. That is the UK side of a report by Reuters, citing the Telegraph, on 11 October that the US had reversed course on its request that developers not share new models with the institute before release.

• Confirmed Narayan told frontier developers: “if you can't adequately control the risks, do not release the system,” and said a breach of UK cyber defences seems “likely, if not inevitable”. GOV.UK speech
• Confirmed Government pilots using frontier AI have identified and addressed more than 1,000 vulnerabilities, and ministers will explore AI-incident reporting through the Cyber Security and Resilience Bill. UK government release, via Wired-Gov
• Confirmed The speech says the AI Security Institute is evaluating the latest models, including Gemini 4 Argon, after working closely with the US government and the labs. GOV.UK speech
• Reported Reuters, citing the Telegraph, reported on 11 October that the US had reversed its request that developers not share new models with the UK institute before release. Reuters, via The Star