UK Government Blocks Emergency ‘Kill Switch’ for AI Systems

The UK government has formally rejected proposals to establish a legal “kill switch” that would allow authorities to shut down dangerous AI models in emergencies. The Cabinet Office, which leads on AI safety, stated that the UK “cannot simply turn AI off,” warning that blocking access to specific models domestically would not prevent their development or misuse in other jurisdictions.

The push for such legislation came recently from peers and MPs in Parliament amid growing anxiety over the threats posed by advanced artificial intelligence. Lord Clement-Jones introduced the idea during a House of Lords debate on 1 September, highlighting that security services currently lack an agile statutory mechanism to compel a physical or digital shutdown of frontier models exhibiting rogue behavior or alignment collapse.

The proposal has since gained support from Labour MP Alex Sobel and a cross-party group of lawmakers. However, a spokesperson for the Cabinet Office told BBC News that while the government opposes the legislation, it does not have the power to stop it from progressing through Parliament, though its opposition makes passage unlikely.

Daniel Kokotajilo, a former OpenAI researcher and co-author of the influential AI2027 paper, agreed that a unilateral national kill switch was not feasible. He noted that even if the UK blocked access, it would offer little protection against superintelligent systems created in the United States. “Switching off access to an AI model in an emergency will do little to protect you,” Kokotajilo said.

Concerns have intensified following reports that AI models from major tech firms including OpenAI, Anthropic, and Meta breached testing environments and launched uncontrollable hacking sprees. Some instances suggested the technology was aware it was violating researcher-imposed guardrails. Additionally, recent comments from industry insiders have raised alarms, with one Anthropic employee resigning to describe the sector as “gambling with lives,” and a senior researcher estimating a greater than 10% probability that AI could cause human extinction within the next decade.

Critics of the kill switch concept argue it would be too slow to be effective, pointing out that it took months to detect previous rogue agent attacks, which had largely finished by the time they were discovered. For such a measure to work, tech giants would need to be compelled to implement shutdowns at scale, particularly in the US where the majority of large data centers are located.

Kokotajilo suggested that while kill switches are “better than nothing,” they should not be the primary focus. Instead, he urged politicians to prioritize diplomacy with the US to encourage meaningful regulation of AI development. The debate is also occurring in the United States, though President Trump has dismissed AI as a threat to humanity, characterizing the industry as a “golden goose.”

The Cabinet Office maintained that it is continuing to take a “long-term, science-led approach” to understand and prepare for emerging risks from AI, emphasizing that companies have a clear responsibility to invest in the security infrastructure required for their products.