UK government rejects an AI 'kill switch,' says it 'cannot simply turn AI off'

[Ivory Molly]

The UK government rejected a proposed legal mechanism that would let it switch off a dangerous AI model in an emergency, with the Cabinet Office telling the BBC that blocking access to models in the UK would not prevent them being developed or misused elsewhere. The kill switch proposal was brought to Parliament by Lord Clement-Jones and a cross-party group including Labour MP Alex Sobel, following growing concern about AI systems exhibiting rogue behavior, compound algorithmic failure, or what Clement-Jones called active alignment collapse

Former OpenAI researcher Daniel Kokotajlo agreed with the government's underlying reasoning, saying switching off access to a model in an emergency would do little to protect anyone, since the same underlying superintelligent systems built elsewhere, largely in the US, would still pose a threat regardless of what Britain does domestically. The Cabinet Office instead emphasized that companies have a clear responsibility to develop their products safely and invest in the security infrastructure this technology requires, framing corporate accountability rather than emergency shutdown powers as the government's preferred approach

The government's opposition doesn't prevent the kill switch bill from continuing through Parliament, but makes it unlikely to actually become law. Critics have argued the harder, less satisfying work, model evaluations that actually stay sealed, mandatory incident reporting, and international coordination that reaches beyond any single country's jurisdiction, matters more than a national off switch that a global model may not even be sitting behind. Curious what people think this decision actually reveals, is the government being realistic about the genuine limits of unilateral national action, or using that realism as cover for avoiding a harder regulatory commitment

UK government rejects 'kill switch' idea for dangerous AIwww.bbc.co.uk

[Dank]

Kokotajlo agreeing with the government's own reasoning here is notable, that's someone who's been genuinely alarmed about AI risk conceding that a national kill switch specifically wouldn't actually solve the underlying problem

[QuantumFoam21]

Worth remembering this doesn't kill the bill entirely, it just makes government support unlikely, individual MPs and Lords can still keep pushing it through the parliamentary process regardless

[Bleak Jason]

Shifting responsibility onto companies to build safely rather than onto government to have shutdown power sounds reasonable but also conveniently avoids Parliament having to actually build any new enforcement infrastructure

[Drogba32]

The UK cannot simply turn AI off line is honestly a refreshingly blunt admission from a government, most political responses to a proposal like this would hedge rather than state the limitation this directly

[Molly4]

International coordination being the harder but more meaningful path is the right diagnosis, but that's also historically been the slowest and least successful part of AI governance to actually achieve in practice