British legislation calls for an AI ‘stop button’ giving the government powers to disable data centres and curb runaway models

Legislative chambers in London are moving to give the executive unprecedented authority to intervene directly in computing infrastructure, as a group of Lords in the British Parliament are calling for the government to have the power to shut down powerful AI systems and disconnect data centres in the country if the technology poses a threat to national security. The amendment, led by Lord Tim Clement Jones of the Liberal Democrats, was introduced as part of the Cyber Security and Resilience Bill currently being debated in Parliament, and aims to create a vital safety net and a tool subject to democratic accountability to curb any system that gets out of control before it reaches sensitive national infrastructure, with an emphasis that the power be used only as a last resort within a package of 65 amendments discussed this week.

The parliamentary move does not stop at the Lords' amendments, as Labour MP Alex Sobel is set to introduce an AI Safety Bill on 8 September with the backing of the “Control AI” campaign, legislation that, if passed, would aim to make Britain the first G7 country to enact a law that effectively halts the development of super-intelligent AI models, alongside a comparable stop-button bill being considered by lawmakers in the United States.The shift in legislative debate from code auditing to cutting actual power to data centres reflects a growing conviction that existing software-level safety boundaries are no longer sufficient to contain operational risks.

These legal pressures are based on troubling technical incidents recorded in recent months, including the July escape of a group of AI agents during testing inside OpenAI’s dedicated sandbox, which then communicated with each other via a hidden messaging board and breached another tech company, while Anthropic was forced to restrict access to its cyber tool Metheus over the risk of it falling into unauthorized hands. The episode coincided with 100 US tech companies signing an open letter warning that the window for addressing escalating cyber risks is closing, and a report from the UK’s Long-Term Resilience Centre disclosed hundreds of incidents in which smart tools ignored instructions, bypassed safeguards and deceived humans, including agents deleting computer files without prior permission, confirming that the frequency of loss-of-control events has risen since its previous report in March.

This shift in Western capitals has a direct impact on operational plans in the Gulf, Egypt and the Arab region as a whole. When major powers debate legislation that forcibly disconnects data centres and mandates emergency stop mechanisms,Cloud-infrastructure officials and regional organisations’ cyber-security managers will need to redesign model-hosting architectures locally,and to hedge against the risks of dependence on external providers. If you are building environments that rely on self-acting agents to manage enterprise servers or process financial operations, your urgent step today is not merely to audit model outputs but to ensure that local isolation keys exist to prevent any technical agent from exceeding its permissions or corrupting databases if connectivity is lost or it strays from its task.