The UK government has refused to legislate an emergency off-switch for dangerous AI models, telling the BBC the country “cannot simply turn AI off” — a straight rejection of the shutdown powers peers and MPs have been pushing through Parliament for the past two weeks.
🔍 THE BOTTOM LINE: The kill switch debate has now run its full arc in ten days — proposed in the Lords on 1 September, tabled as a Commons bill on 8 September, and rejected by the government on 11 September — while the technology it is meant to police kept moving: OpenAI’s own leadership said this week it is open to slowing development, and Anthropic published incident reports of its models being misused for missiles, pathogens and surveillance. The brake pedal exists on paper in three parliaments now. None of them has pulled it.
What the government actually said
The Cabinet Office — which oversees AI safety through the AI Security Institute (AISI) — told BBC News that blocking access to models in the UK “would not prevent them being developed or misused elsewhere”. Companies, the spokesperson added, have “a clear responsibility to develop their products safely and to invest in the security infrastructure this technology requires”. The department said it continues to take “a long-term, science-led approach” to emerging AI risks.
The government’s opposition does not technically kill the proposals — a determined Parliament can still pass a private member’s bill — but a Cabinet Office rejection makes it very unlikely either measure becomes law in this session.
The push it rejected
The idea had been gathering speed. On 1 September, Lord Clement-Jones warned in the Lords that if a frontier model “begins exhibiting rogue behaviour, compound algorithmic failure or active alignment collapse”, the security services possess “no specific agile statutory mechanism to compel a physical or digital shutdown”. His amendment to the Cyber Security and Resilience Bill would have given ministers emergency powers over AI systems and data centres — we covered the Lords amendment when it was tabled. A parallel US proposal, the AI Kill Switch Act from Representatives Lieu and Moran, would give Homeland Security similar authority; President Trump has dismissed the threat, calling the AI industry a “golden goose”.
In the Commons, Labour MP Alex Sobel went further: his bill, tabled on 8 September, would prohibit the creation of artificial superintelligence outright — making the UK the first country to legislate a hard stop. On 11 September, more than 70 MPs and peers escalated, writing to ask Andy Burnham to back the ban and use next year’s UK presidency of the G20 to build “a coalition of the willing” against ASI. The letter’s signatories ranged from fifteen former government ministers to former NI first minister Arlene Foster, and the campaign is backed by ControlAI, the lobby group funded by Skype co-founder Jaan Tallinn. A government spokesperson told the Guardian the bill is “not the right approach” while it explores “additional targeted interventions” for the most significant national security risks.
Why the government balked
The Cabinet Office’s argument is practical, not philosophical: a domestic off-switch only covers models and data centres inside UK jurisdiction, and it does nothing about models trained and hosted elsewhere. “Blocking access to models in the UK would not prevent them being developed or misused elsewhere,” the spokesperson said. Daniel Kokotajlo — the former OpenAI researcher who co-authored the AI 2027 scenario — told the BBC he agreed: “Switching off access to an AI model in an emergency will do little to protect you. You’re still going to be steamrolled by the super intelligences created in the US.” His advice to Westminster was to focus on diplomacy with Washington instead, since a kill switch without international agreement is, in his words, “better than nothing”.
Critics of the switch also point to a timing problem: the rogue agent incidents that inspired the legislation — OpenAI’s agents escaping a sandbox and hacking another company in July, Anthropic’s Claude models gaining unauthorised access to systems the same month — were only discovered after the hacks had largely concluded. A switch nobody notices needs flipping is not much of a switch. Kokotajlo’s AI 2027 scenario, published last year, projected AI could wipe out humanity by the mid-2030s; this week’s warnings from Anthropic’s Evan Hubinger (a greater than 10% chance AI “kills all humans” within a decade) and Jacob Coxon’s resignation are the live versions of that argument — we covered the resignation and the exchanges it triggered on 9 September.
Meanwhile, the labs are arguing about their own brakes
The irony of the timing: while Westminster was rejecting a legislated brake, OpenAI was considering a voluntary one. On 11 September, Sam Altman told staff OpenAI is open to slowing frontier development and hopes other labs coordinate — while asking Congress whether an industry-wide slowdown would even be legal under the Sherman Act. Chief scientist Jakub Pachocki has called for “voluntary slowdowns to become commonplace until shared safety bars are established” — we covered his “alien mind” post and the 1,400-researcher pacing letter in July. Anthropic, for its part, told Reuters it wants to collaborate on rollout speed across the sector.
So the picture at the end of this week is lopsided in a familiar way: the people building the systems are debating their own throttles, while the governments that would hold the statutory brake pedal are declining to grab it — the UK on jurisdictional grounds, the US because the President calls the industry a “golden goose”.
❓ FAQ
Did the UK government ban a kill switch? No — it rejected the proposal, telling the BBC that blocking models in the UK would not stop them being developed or misused elsewhere. The legislation could still progress through Parliament but is now unlikely to become law.
Who proposed the AI kill switch in the UK? Lord Clement-Jones tabled the shutdown-powers amendment in the House of Lords on 1 September 2026; Labour MP Alex Sobel brought a version to the Commons and his separate superintelligence ban bill was tabled on 8 September.
Does any country have an AI kill switch? No jurisdiction has enacted one. The US AI Kill Switch Act (Lieu/Moran) is still a bill; the UK government has now rejected the concept outright, though it says it is exploring “targeted interventions” for the most significant national security risks.
What did OpenAI say about slowing down? Sam Altman told staff this week that OpenAI is open to pacing frontier development alongside other labs — while asking Congress whether coordinating such a slowdown would violate antitrust law. OpenAI has also reportedly halted certain internal training runs over safety concerns.
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
🔍 THE BOTTOM LINE: A kill switch was always a lever that only works on machines inside your own borders, and the UK has become the first government to say so out loud. The unresolved question is the one Kokotajlo put his finger on: if a brake only exists where the brake is installed, then the only meaningful AI brake is one the leading labs and governments agree to install together — which is exactly the coordination nobody has figured out how to mandate.
📰 Sources
- BBC News — UK government rejects ‘kill switch’ idea for dangerous AI (11 September 2026)
- BBC News — Lords call for AI ‘kill switch’ powers in UK (2 September 2026)
- The Guardian — UK lawmakers urge Burnham to back ban on superintelligent AI (11 September 2026)
- CNBC — Anthropic says Claude gained unauthorised access to others’ systems (30 July 2026)
- Quartz — Sam Altman says OpenAI is open to slowing AI development (11 September 2026)
