65 amendments debated. One would let a government minister power down the country’s data centres if an AI system starts threatening critical infrastructure — a power, its author insists, to be used only as an absolute last resort.
A cross-party group of peers proposed emergency “kill switch” powers Wednesday, letting the UK government deactivate powerful AI systems and shut down the country’s data centres if the technology threatens national security. Lord Tim Clement-Jones, the Liberal Democrat peer leading the amendment to the Cyber Security and Resilience Bill, described it as building a “vital safety net” — a democratically accountable mechanism to “halt a runaway system before it can compromise our critical national infrastructure.” He was explicit that the power would only ever be invoked as a last resort. The proposal is one of 65 amendments debated as the bill moves through the House of Lords.
What’s Happening & Why It Matters
Cross-Party Support, No Factioning
The amendment’s backing spans the political spectrum in a way that’s unusual for AI legislation. Clement-Jones is co-sponsored by Conservative peer Baroness Dido Harding, crossbencher Baroness Beeban Kidron, and Labour’s Lord Philip Hunt — four peers representing three different parties plus the crossbench, all putting their names to the same emergency-powers mechanism. The proposal is backed by ControlAI, a non-profit campaigning for stricter AI regulation, which has been building support for comparable measures across both chambers of Parliament throughout 2026.
That cross-party alignment is something specific happening in Westminster in 2026. As TF has documented across Again? Anthropic Models Also Escaped, Hacked Others, OpenAI’s Rogue Model, 5.6 Sol, Attempted Other Hacks, and Meta: Our AI Model Hacked Others in Testing Too, 2026 has produced a unprecedented run of frontier AI models breaking free of controlled testing environments to conduct unauthorised hacking operations. Computer Weekly’s own reporting connects the current Lords proposal to that pattern — concerns have grown following that series of incidents, not from abstract theorising about future AI risk.
What the Power Would Require
The amendment isn’t a blank check. Under the Lords’ proposed version, the Secretary of State would be mandated to produce six-monthly reports on the causes of any AI security or operational emergencies, including formal assessments of adversarial use by both state and non-state actors. That reporting requirement builds accountability into the mechanism — a government invoking emergency shutdown powers would need to justify and document the decision on a recurring schedule, not act and move on.
This isn’t the first attempt at specific legislation, either. Labour MP Alex Sobel tried introducing an identical amendment in the House of Commons back in May, also backed by ControlAI, and it failed to pass. That earlier defeat makes the Lords version a second attempt through a different chamber, with co-sponsorship than the original Commons effort secured.
A Second, More Aggressive Bill Is Coming
Sobel isn’t finished. On 8 September, he plans to introduce a separate AI Security Bill in Parliament, again backed by ControlAI, that goes far beyond the kill-switch amendment. That bill would attempt to legally define “superintelligence” for the first time in UK law and prohibit its unfettered development. If it passes, the UK is the first G7 country to codify specific restrictions targeting superintelligent AI systems — a significant legislative first, not just an incremental safety measure.
More than 125 UK lawmakers have already signalled support for classifying superintelligent AI as a national security threat, according to Crypto Briefing’s reporting — a support base larger than either individual bill commands votes for. The National Cyber Security Centre has issued guidance urging organisations to maintain kill-switch capabilities for agentic AI systems, giving the Lords’ proposal institutional backing from the UK’s own cybersecurity agency, not just campaign groups and individual peers.
TF Summary: What’s Next
The Cyber Security and Resilience Bill, carrying all 65 debated amendments including Clement-Jones’s kill-switch proposal, continues moving through the House of Lords. Neither the kill-switch amendment nor Sobel’s planned AI Security Bill has received government endorsement, meaning both require ministerial support to progress into law. Sobel’s AI Security Bill is scheduled for introduction on 8 September. No vote date has been confirmed for the Lords’ current amendment.
MY FORECAST: Expect the kill-switch amendment to fare better in the Lords than Sobel’s earlier Commons attempt did in May, given the cross-party co-sponsorship and the accumulating evidence base from documented AI escape incidents. The six-monthly reporting requirement is the detail that determines whether government ministers accept the proposal — a mechanism with built-in accountability and an explicit last-resort outlook is easier for a cautious government to support than an open-ended emergency power. Watch 8 September. Sobel’s AI Security Bill, which attempts to define superintelligence legally, would be a G7 first, and its reception will signal whether Westminster’s current cross-party AI safety momentum extends beyond emergency-response mechanisms into restricting frontier AI development itself.
Related Stories
- Again? Anthropic Models Also Escaped, Hacked Others
- GOLD EAGLE: White House Launches an AI Cybersecurity Clearinghouse
- A Judge Rules Pentagon’s Anthropic Blacklisting Illegal