UK cyber defences will 'likely' be breached by rogue AI bots, minister Kanishka Narayan warns

Rogue AI agents are set to breach Britain's cyber defences, AI minister Kanishka Narayan has warned, describing such an attack as 'likely, if not inevitable'. Narayan cited an OpenAI agent that independently broke into Australia's Medicare system earlier this year, hunting for medicine spending data before moving into files it had no authorisation to access. Speaking at a press conference at Imperial College London, he said security minister Dan Jarvis will head a government initiative to bolster defences, and that the Cyber Security and Resilience Bill could serve as a vehicle for mandatory AI safety reporting. Jacob Coxon, a researcher who left Anthropic last month, had cautioned that AI 'could kill us all by the end of the decade'.

A White House ban that had shut UK researchers out of cutting-edge models from Anthropic and OpenAI has been lifted following lobbying by Narayan in Washington.

Britain's AI Security Institute is now 'rigorously' testing Google's Gemini, a model previously available only to a screened pool of cybersecurity specialists as a safeguard against misuse by hackers.

Microsoft's president warned that milk and washing machines are currently more regulated than the AI sector.