Britain is set to expand its cyber-defence capabilities while urging companies developing the most advanced artificial-intelligence systems to take greater responsibility for their security. AI minister Kanishka Narayan will argue that the UK cannot depend entirely on technology companies to police their own models and will call for more domestically controlled AI infrastructure.
The intervention comes amid warnings that hostile states could steal unreleased frontier models. Geoffrey Irving, a former chief scientist at the UK’s AI Security Institute who has also worked at OpenAI and Google DeepMind, has said existing protections may not be sufficient to prevent an adversarial government from obtaining advanced systems. He pointed to an OpenAI security-testing programme in which researchers found that the potential scope of access was extensive. OpenAI says it uses safeguards including encryption and continuous monitoring.
The government’s wider cyber programme includes plans for AI-powered systems that could identify and address vulnerabilities across critical networks at a speed beyond human teams. Ministers have also called on leading AI companies to work directly with the government on national cyber-defence tools. A voluntary business pledge calls for cyber security to be treated as a board-level responsibility, registration with the National Cyber Security Centre’s Early Warning service and Cyber Essentials requirements for supply chains. The government has committed £90 million over three years to improve resilience, including among small and medium-sized businesses.
The minister’s portfolio includes AI policy, data infrastructure, the UK AI research resource, the AI Security Institute and Sovereign AI. Microsoft chief executive Satya Nadella has separately argued that advanced models should operate inside contained systems with observable behaviour, testable limits and a human-controlled emergency stop. Together, these developments put security, national control and safeguards alongside capability as central tests for Britain’s AI strategy.