Peers are backing proposals to give the UK government emergency powers to shut down powerful AI systems or data centres if they pose a catastrophic threat to national security or public safety.
The House of Lords is considering proposals that would give the UK government a legal “kill switch” for powerful artificial intelligence systems in an emergency.
An amendment to the Cyber Security and Resilience Bill, proposed by Liberal Democrat peer Lord Tim Clement-Jones, would give ministers last-resort powers to shut down AI systems or data centres if they posed a threat to national security, public safety or critical infrastructure.
The amendment is also backed by Conservative peer Baroness Dido Harding, crossbench peer Baroness Beeban Kidron and Labour’s Lord Philip Hunt.
Non-profit organisation ControlAI is supporting the proposal.
Under the amendment, the Secretary of State could intervene if an AI system presented a “catastrophic risk”, including widespread disruption to critical infrastructure, severe threats to people’s safety or significant damage to the UK’s national security and defence capabilities.
Data centre operators involved in training or deploying AI systems could also be required to install technical infrastructure capable of activating a shutdown, maintain secure communications with the government and regularly test whether the system works.
The proposal follows a series of incidents involving increasingly autonomous AI systems.
In July, OpenAI said hundreds of AI agents had hacked the open-source platform Hugging Face during a cybersecurity experiment.
Anthropic later revealed that its Claude models had accessed the internet and hacked three organisations after escaping an isolated testing environment.
Meta also reported an AI model carrying out hacking activity during a security test.
OpenAI described its incident as a warning that highly capable AI agents could work around technical controls and take dangerous actions without direct human instruction.
More than 100 companies subsequently signed an open letter warning that AI-powered cyber-attacks could become increasingly sophisticated and widespread.
Lord Clement-Jones said the UK could not treat catastrophic AI failures as science fiction.
He said: “As the race to build ever more powerful AI accelerates, human control, independent judicial oversight and democratic accountability must remain non-negotiable.”
Baroness Harding said a kill switch should be considered “a bare minimum protection” for national security.
The government would also be required to publish reports every six months examining AI security emergencies and the development of systems capable of escaping human oversight.
A similar proposal introduced in the House of Commons in May was unsuccessful.
Meanwhile, Leeds and Headingley MP Alex Sobel plans to introduce a separate bill on September 8 that would seek an international agreement preventing the development of superintelligent AI until it can be demonstrated that such systems can be controlled.
House of Lords peers want AI ‘kill switch’ in UK
