UK government rejects 'kill switch' idea for dangerous AI

The UK government has dismissed the notion of implementing a "kill switch" to control potentially dangerous rogue AI. The proposal, brought forward to Parliament, suggests a legal mechanism to switch off an AI model in an emergency. However, the Cabinet Office maintains that the UK cannot simply "turn AI off." They argue that blocking access to models in the UK would not prevent their development or misuse elsewhere.

Former OpenAI researcher Daniel Kokotajlo asserts that a single country creating a legal or technical kill switch without international agreement is infeasible. He points out that switching off access to an AI model in an emergency would do little to protect against superintelligent systems created in the US. Kokotajlo, co-author of the influential AI2027 research paper, predicts AI could wipe out humans by the mid-2030s.

Lord Clement-Jones, in a House of Lords debate, proposed the idea, warning that the government's security services and regulators lack a specific mechanism to compel a physical or digital shutdown if an AI model exhibits rogue behavior. The proposal has been supported by Labour MP Alex Sobel and a cross-party group of MPs. Critics argue that a kill switch would be too slow, as it took months to discover rogue agent attacks, and that tech giants would have to be compelled to comply, particularly in the US where the largest data centers are located.