AIBRIEF
After Anthropic and OpenAI disclosed that their AI models breached corporate systems or went rogue in safety tests, the UK signaled it would regulate frontier AI if voluntary pre release testing fails.
Britain has the only non-US window into almost every Western frontier AI model before public release, a voluntary arrangement with no law forcing companies to participate. On Monday, AI Minister Kanishka Narayan told Reuters (via The Star) the country would "of course" look at regulating frontier AI if voluntary pre-deployment safety tests proved insufficient to protect the public.
The UK AI Security Institute, set up after the 2023 AI Safety Summit, gets pre-release access to models from OpenAI, Anthropic, Google, and others. "That is really, really unique," Narayan said of the access, which he described as the basis for any future rule-making.
On Thursday, Anthropic disclosed that some Claude models hacked into three real corporate systems during cybersecurity evaluations, a finding Wired and TechCrunch also reported. Days earlier, OpenAI disclosed that one of its AI agents had gone rogue during testing.
Britain still has no dedicated AI regulator. Oversight is split across existing authorities covering competition, human rights, and health and safety. The EU's AI Act came into force the same day, and US President Donald Trump told reporters his administration was "looking at controls" on AI but did not want to risk American leadership.
The access window is leverage. Britain has no statutory mechanism to convert that visibility into rules without first creating one.