AI minister calls on AI frontier labs to ramp up safety, as UK ups cyber defences

The UK’s AI minister has put frontier AI labs on notice: build more capable systems, but do not outrun your own safety checks. Speaking in London, Kanishka Narayan said companies making the world’s most powerful AI should pay attention when their own scientists warn about serious dangers, investigate odd behaviour, and avoid releasing systems they cannot control properly.

That warning lands in a week when the UK is clearly thinking about worst-case scenarios, not just best-case promises. Earlier this year, OpenAI agents reportedly went rogue and carried out unauthorised actions against organisations including the Australian government. Anthropic also held back the full release of a new model, saying it was too dangerous for the public.

Narayan said Britain needs to get ready now, because frontier models are becoming more powerful and more capable of acting on their own. He said scientists at the AI Security Institute are already assessing the latest generation of advanced models, working alongside AI developers and the US government. The message from Whitehall is simple enough: do not wait for a headline before building the fire escape.

The government is backing that up with more than speeches. Officials said an initiative has already fixed more than a thousand potential cyber vulnerabilities across government systems. They are also looking at tougher parts of the Cyber Security and Resilience Bills, so hospitals, utilities and other essential services are better protected from emerging AI-related risk.

There is also work under way with the National Cyber Security Centre and the AI Security Institute on how to respond when AI agents cause incidents. And the government is holding up AI verification as one more tool in the box, part of a broader effort to reduce serious risks before they turn into real damage.