A group of UK peers wants the government to have a big red button for AI, and they’re not being metaphorical about it.
Lord Tim Clement-Jones, the Liberal Democrat peer, is leading an effort to amend the Cyber Security and Resilience Bill with provisions that would grant emergency powers to deactivate advanced AI systems and shut down data centres if they’re deemed a threat to national security.
What the proposals actually involve
The amendments being debated in the House of Lords would create a legal framework for government intervention when AI systems cross from “useful tool” into “existential concern” territory. The power wouldn’t just cover individual AI models. It extends to the physical infrastructure powering them, meaning entire data centres could be ordered offline if the situation warranted it.
The timing is deliberate. On September 8, 2026, Labour MP Alex Sobel plans to introduce a separate AI Security Bill, backed by the campaign group ControlAI. That bill takes things a step further by attempting to define “superintelligence” in legal terms for the first time and would prohibit its unfettered development.
If enacted, the UK would become the first G7 country to codify restrictions specifically targeting superintelligent AI systems. Over 125 UK lawmakers have already signaled support for classifying superintelligent AI as a national security threat.
The National Cyber Security Centre has also weighed in, issuing guidance urging organizations to maintain kill-switch capabilities for agentic AI systems. Agentic AI refers to systems that can take actions autonomously, making decisions and executing tasks without human approval at every step.
The government’s position: polite reluctance
Despite the parliamentary momentum, the UK government has so far preferred a lighter touch. The current stance favors voluntary safeguards over embedding kill-switch powers directly into legislation.
Lord Clement-Jones and his allies appear unconvinced by the voluntary approach. Their argument rests on a straightforward premise: if an AI system poses a genuine national security risk, the government needs legal authority to act immediately, not the ability to send a strongly worded letter requesting voluntary compliance.
A transatlantic pattern emerges
The UK isn’t operating in isolation. On July 23, 2026, US legislators introduced the AI Kill Switch Act, which would impose potential fines of $20M per day for non-compliance.
Both sets of proposals currently await formal government backing to advance through their respective legislative processes. The broader context includes earlier House of Lords discussions about autonomous weapons systems and the oversight gaps that emerge when machines can make consequential decisions faster than humans can review them.