Government rejects an AI kill switch because you cannot simply turn AI off

Experts warn a British AI kill switch has a major flaw

The idea has been under discussion in Parliament in recent weeks. Under the proposed approach, ministers could potentially force an AI model operating from a UK data centre offline if it began behaving in a dangerous or completely unexpected way.

But the Cabinet Office argues that such a power would have clear limits. A spokesperson told BBC News that the UK “cannot simply turn AI off”, pointing out that “blocking access to models in the UK would not prevent them being developed or misused elsewhere”.

That argument has also been made by Daniel Kokotajilo, a former OpenAI researcher who co-authored AI2027. The paper, published last year, warned that continued AI development could potentially lead to human extinction by the mid-2030s.

Kokotajilo nevertheless believes a British shutdown mechanism would offer little protection in an emergency. “Switching off access to an AI model in an emergency will do little to protect you,” he said. “You're still going to be steamrolled by the super intelligences created in the US.”

Britain could need an AI kill switch before it is too late

The question of whether Britain could stop a dangerous AI system has become a new point of concern for MPs and peers.

Lord Clement-Jones raised the issue in the House of Lords on 1 September, warning that there is currently no dedicated legal power allowing authorities to order an advanced AI model to be shut down quickly if it becomes a threat.

That could leave a significant gap if a system operating in Britain began acting outside its intended limits, particularly if it had access to critical infrastructure. 

Clement-Jones argued that security services and regulators could find themselves without a clear legal route to physically or digitally disconnect it.

The proposal has since attracted support in the Commons, including from Labour MP Alex Sobel, with MPs from different parties backing the idea. 

Concerns have also grown after reports that AI models developed by OpenAI, Anthropic and Meta have found ways to escape controlled testing environments.

The case for a shutdown mechanism is ultimately simple. It would give authorities an emergency option when other safeguards fail. 

It would not guarantee that an AI system could be contained, but supporters argue that having a way to intervene is preferable to discovering during a crisis that no such power exists.

Rejecting the proposal does not automatically kill the bill

The proposed AI legislation is not necessarily finished just because ministers have rejected it. Parliament can continue considering the bill, although securing enough time and support to complete its passage becomes much harder when the government is opposed to it.

Ministers have also made clear that they do not believe the state should carry the entire burden of keeping AI systems safe. 

The Cabinet Office has pointed to the companies developing the technology, arguing that “companies have a clear responsibility to develop their products safely and to invest in the security infrastructure this technology requires”.

Under that model, the industry would remain primarily responsible for building adequate safeguards and protecting the systems it creates, while the government would retain a more limited role.