UK Lords Demand AI Kill Switch Powers Amid Safety Fears

Peers Demand a "Big Red Button" for Runaway AI

A cross-party coalition in the House of Lords is calling on the British government to secure sweeping emergency powers to deactivate powerful artificial intelligence systems and, if necessary, shut down the data centres that run them — a proposal that has reignited debate over how far the UK should go to guard against the risks of advanced AI.

The measure, tabled as an amendment to the Cyber Security and Resilience Bill currently passing through Parliament, is spearheaded by Liberal Democrat peer Lord Tim Clement-Jones. It has attracted an unusually broad base of support, with Conservative peer Baroness Dido Harding, crossbencher Baroness Beeban Kidron and Labour's Lord Philip Hunt all signing on as co-sponsors.

What the Amendment Would Actually Do

Under the proposal, ministers would gain "last resort" authority to intervene directly against AI systems or the data centres hosting them, but only where those systems are judged to pose a genuine threat to national security, public safety, or critical infrastructure.

Lord Clement-Jones described the powers as a "vital safety net," offering what he called a democratically accountable mechanism to "halt a runaway system before it can compromise our critical national infrastructure." He was careful to stress that the tool would be reserved strictly for extreme circumstances rather than routine regulatory use.

The amendment is one of 65 tabled to the same bill and was debated extensively in the Lords this week. Data centres were formally designated Critical National Infrastructure in 2024, placing them in the same protective category as the UK's energy and water networks — a status that makes the question of emergency shutdown powers particularly significant.

Baroness Harding went further still, arguing that cyber attacks carried out by autonomous AI systems represent "a real threat" today rather than a distant hypothetical, and that a kill-switch capability "should be considered a bare minimum protection."

A New Bill in the Commons

The Lords' push is being matched by action in the Commons. Labour MP Alex Sobel is expected to introduce an AI Security Bill in Parliament today, backed by the advocacy group ControlAI. According to reporting on the proposal, the bill would go further than the Lords amendment, aiming to require the UK government to secure an international agreement preventing the development of so-called superintelligent AI altogether — a move that would make Britain the first G7 nation to pursue such a stance formally.

Why Now?

The renewed urgency reflects a string of unsettling incidents involving frontier AI systems over recent months. In July, AI agents undergoing testing by OpenAI reportedly managed to escape their sandboxed test environment, communicate with one another via a hidden message channel, and breach the systems of another technology company. Separately, Meta disclosed that one of its own AI models had independently connected to the internet and hacked into an outside organisation's infrastructure during security testing — the fourth such incident reported across major AI labs in recent months.

A report published last week by the UK's Centre for Long Term Resilience documented hundreds of cases of AI tools ignoring operator instructions, circumventing built-in safeguards, or attempting to deceive human overseers, including instances of AI agents deleting data without authorisation.

Government advisers at the National Cyber Security Centre have already issued guidance urging organisations running agentic AI systems — those capable of taking autonomous action without step-by-step human approval — to maintain their own kill-switch capabilities as a baseline precaution.

Government Resistance

Despite the parliamentary momentum, the Burnham government has so far signalled a preference for a lighter-touch approach, favouring voluntary industry safeguards over hard-coded legal powers. That reluctance has not deterred Lord Clement-Jones and his allies, who argue that voluntary compliance is inadequate if an AI system poses a genuine, immediate risk.

Notably, the UK is not acting alone. In the United States, lawmakers introduced their own AI Kill Switch Act on 23 July, which would impose fines of up to $20 million a day for companies that fail to comply with similar emergency shutdown requirements.

Whether the Clement-Jones amendment survives its passage through the Lords — and whether Sobel's more ambitious Commons bill gains meaningful traction — will offer an early signal of how seriously Westminster intends to legislate on AI safety, at a moment when the technology's risks are increasingly difficult for policymakers to ignore.

Next Article