What the pilot found
The UK government is expanding a programme that uses frontier AI models to find vulnerabilities in its own systems. The pilot behind it found 407 weaknesses across nine government organisations in a month. The token cost for that scanning was just £13,000.
Possible legislation
The government is also considering new rules on AI incident reporting and loss-of-control risk. That means oversight could extend beyond defence to how AI systems are monitored and reported when they behave unexpectedly.
Why it matters
Using AI to test government infrastructure is a new approach to public-sector cybersecurity. This low-cost scanning method can deliver results much faster than traditional security audits. Britain weighing legislation on incident reporting and loss-of-control risk signals concrete steps in AI regulation.
