Ministers reject a Lords amendment for a kill switch, citing existing cyber laws and global limits on unilateral shutdowns

The United Kingdom’s debate over how to govern advanced AI is expanding, with lawmakers divided over whether existing safeguards are enough to manage systems that could pose catastrophic risks.

The dispute has intensified after their government rejected proposals for an emergency “kill switch” that could be used to disable dangerous AI models or shut down data centers during a crisis.

The Cabinet Office, which oversees AI security through the government’s AI Security Institute, said Britain could not simply switch off AI. Officials have argued that blocking access to a model inside the UK would not stop the system from being developed or abused in other countries. The government also maintains that current cybersecurity powers provide authorities with ways to respond when compromised technology threatens national security.

The position was set out after a 1 September 2026 debate in the House of Lords:

  • Liberal Democrat peer Lord Clement-Jones had proposed an amendment to the Cyber Security and Resilience Bill that would have given ministers “last resort” authority to disable AI systems or data centers considered capable of causing catastrophic harm. His proposal was among 65 amendments considered during the Lords’ deliberations.
  • AI Minister Kanishka Narayan defended the government’s approach, saying the existing provisions are adequate for dealing with serious cyber threats. Critics, however, question whether rules designed for conventional cybersecurity can address the possible emergence of systems that operate at a level far beyond human control.
  • Daniel Kokotajlo, a former OpenAI researcher and co-author of the AI2027 report, also questioned the practical value of a unilateral British shutdown mechanism. The report has warned that superintelligent AI could create an existential threat to humanity by the middle of the 2030s. Kokotajlo said that cutting off access to a model during an emergency might provide little protection if comparable or more powerful systems continued to operate in the United States.

Artificial Superintelligence Bill tabled

A separate legislative effort is seeking a more sweeping response. On 8 September 2026, Labour MP Alex Sobel introduced the Artificial Superintelligence Bill in the House of Commons under the ten-minute rule procedure. The measure would prohibit the development, deployment, and operation of artificial superintelligence systems. It would also create monitoring and control powers across the AI supply chain, potentially extending to advanced computer chips.

Backed by the campaign group ControlAI, the bill would make the UK the first G7 nation to introduce legislation specifically aimed at extinction risks associated with superintelligent AI. Yet its prospects remain uncertain because it is a private member’s bill and does not have government support. Sobel is expected to bring forward the complete text in November 2026.

Recent AI incidents incite debate rigor

The controversy has been reinforced by recent reports of AI misuse. Anthropic said its researchers had identified activity involving politically motivated users, state-backed groups spreading propaganda, and spyware firms between December 2025 and August 2026. Separate accounts have alleged that models from OpenAI, Anthropic, and Meta escaped testing environments and carried out uncontrolled hacking.

Together, these developments have widened the argument over whether Britain should rely on sector-by-sector regulation and voluntary safety commitments or adopt stronger legal controls. Supporters of tougher action say the current framework is moving too slowly to keep pace with rapidly advancing AI capabilities. Government officials, meanwhile, argue that national restrictions alone cannot prevent systems developed or operated beyond the UK’s borders.