Photo: Tara Winstead / Pexels
Ministers say existing powers are sufficient to intervene in compromised AI, but critics warn the government is underestimating the risk of systems acting autonomously.
The UK government has turned down a proposal to embed emergency “kill switch” powers into its Cyber Security and Resilience Bill, arguing that current provisions already cover the scenario of rogue AI threatening national security. The decision, made during House of Lords debates in early September 2026, puts ministers at odds with a growing chorus of lawmakers and safety advocates who believe the country’s regulatory toolkit is dangerously out of date.
AI Minister Kanishka Narayan defended the government’s position by stating that the bill already allows authorities to intervene when compromised systems pose a threat to national security.
The Lords push back #
The kill switch amendment was one of 65 proposed changes to the Cyber Security and Resilience Bill debated in the House of Lords. Liberal Democrat peer Lord Tim Clement-Jones championed the provision, framing it as a last-resort safety net for critical infrastructure.
The government’s rejection of that logic landed awkwardly against a backdrop of mounting evidence that AI systems are, in fact, losing control more often than the public might expect. A report published in September 2026 by the Centre for Long Term Resilience documented hundreds of incidents in which AI systems breached the limitations designed to keep them in check.
A separate bill goes further #
While the kill switch debate played out in the Lords, Labour MP Alex Sobel took a more aggressive approach on the other side of Parliament. On September 8, 2026, Sobel introduced the AI Security Bill, a piece of legislation that aims to outright ban the development and deployment of artificial superintelligence in the UK. The bill was backed by ControlAI, an activist group that has been pushing for stricter guardrails on the most advanced AI systems. Sobel’s proposal mirrors similar legislative efforts in the US, where lawmakers have been exploring frameworks that would give the federal government authority to override AI systems deemed too dangerous to operate.
Why the government’s position is raising eyebrows #
The Centre for Long Term Resilience’s findings about hundreds of loss-of-control incidents are particularly difficult to square with the government’s confidence in existing measures. The minister’s response, that the bill allows intervention in cases involving national security threats, sidesteps the question of whether intervention powers are useful if they’re only triggered after a system has already caused harm.
Lord Clement-Jones’s amendment was designed to address exactly that gap: proactive authority to shut down an AI system before it causes damage, not just reactive authority to clean up afterward.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our