UK government rejects 'kill switch' idea for dangerous AI

The UK government has pushed back on a plan to give ministers a legal emergency switch for dangerous AI. The idea has been floated in Parliament by peers and MPs as fears grow about rogue models, but the Cabinet Office says the country “cannot simply turn AI off.”

Its basic objection is blunt: even if Britain blocked access inside the UK, the model could still be developed or misused somewhere else. That leaves the proposal looking more symbolic than practical, and the government’s stance makes it unlikely to become law, even though it can still keep moving through Parliament.

The debate has been driven by worries about highly capable autonomous systems and the possibility of a model going off the rails in a data centre or inside critical infrastructure. In the Lords on 1 September, Lord Clement-Jones argued that regulators and security services currently lack a specific statutory tool to force a shutdown if a frontier model starts showing rogue behaviour, compound algorithmic failure or “active alignment collapse.” Labour MP Alex Sobel later raised the same idea in the Commons, with cross-party backing.

Daniel Kokotajlo, a former OpenAI researcher, backed the government’s basic point that a single country can’t realistically create a kill switch without international agreement. He said cutting access in one place would do little if more powerful systems were built elsewhere, and he pointed to the risk of being “steamrolled” by superintelligences created in the US.

The Cabinet Office also said companies have a clear responsibility to build products safely and invest in the security infrastructure this technology needs. That sits closer to the real argument here: whether governments should plan for a dramatic off switch, or force companies to make the thing less likely to need one in the first place.