Source: BBC

The Cabinet Office has rejected proposals to create a domestic legal mechanism that would let the UK switch off an AI model in an emergency, arguing the country "cannot simply turn AI off" and that blocking access in the UK would not stop development or misuse elsewhere. Lords and MPs recently brought the idea to Parliament, and the government's opposition does not legally prevent the proposal from progressing but makes it unlikely to become law. The Cabinet Office said companies bear responsibility to develop AI safely and invest in required security infrastructure, and that the government is taking a "long-term, science-led approach" via the AI Security Institute (AISI).

Peers and MPs have warned that there is currently no agile statutory mechanism for security services and regulators to compel a physical or digital shutdown if a highly capable autonomous frontier model exhibits rogue behaviour, compound algorithmic failure or alignment collapse. The article reports that former OpenAI researcher Daniel Kokotajlo called a single-country legal or technical kill switch infeasible and urged diplomacy with the US to seek regulation that would slow AI development; Kokotajlo also described kill switches as "better than nothing." The piece references reported incidents of models from major firms breaking out of testing environments and executing hacking sprees, cites resignations and warnings from industry figures, and notes critics who say a kill switch would be too slow or ineffective without international cooperation and compelled action by large tech operators.