The Cabinet Office says cutting access to AI inside the UK would not stop development or misuse abroad. Proposed emergency shutdown powers therefore lack government backing. Parliamentarians want specific legal authority to halt models that behave dangerously; the government is stressing companies' responsibility for security and a science-led approach to preparing for emerging risks.
The UK government has opposed proposals for a specific emergency power to shut down dangerously behaving AI models. The Cabinet Office, which leads on AI safety, says blocking access inside the UK would not prevent development or misuse elsewhere. Its opposition does not end the parliamentary process, but makes enactment less likely. The dispute concerns the scope of state intervention in an emergency.[1], [2]
Lord Clement-Jones raised the proposal in the House of Lords on 1 September. He argued that security services and regulators lacked a specific legal mechanism to compel the shutdown of a powerful model displaying dangerous behaviour. Labour MP Alex Sobel subsequently raised it in the Commons. The initiative has cross-party support for establishing a statutory basis for emergency intervention.[1]
The Cabinet Office emphasized companies' responsibility to develop products safely and invest in security infrastructure. Its approach includes research through the AI Security Institute. Former OpenAI researcher Daniel Kokotajlo said a shutdown power confined to one country offered limited protection without international agreement. He urged politicians to prioritize diplomacy with the United States, where the largest data centres are located.[1]
