Lords call for AI 'kill switch' powers in UK
Its backers say it would provide a "vital safety net" against runaway AI systems such as those from OpenAI and Anthropic.
Peers in the UK are advocating for the British government to possess the authority to disable potent AI systems and power down the nation's data centers if these technologies jeopardize national security.
This initiative is spearheaded by Lord Tim Clement-Jones of the Liberal Democrats, who has put forward this proposal as an amendment to the Cyber Security and Resilience Bill, currently under parliamentary review. Lord Clement-Jones stated that this measure would establish a "vital safety net" and offer a democratically accountable method to "halt a runaway system before it can compromise our critical national infrastructure." He emphasized that this tool would only be deployed as a final option.
This is one of 65 amendments to the same bill that underwent debate this week.
Concurrently, on September 8, Labour MP Alex Sobel intends to introduce an AI Security Bill in Parliament, backed by the campaign group ControlAI. If enacted, this bill would position the UK as the first G7 nation to implement legislation that would effectively pause the advancement of superintelligent AI. Both proposals would ultimately necessitate government endorsement to proceed.
An AI Kill Switch Act is also being considered by legislators in the United States.
Recent weeks have seen heightened examination of AI's impact on cybersecurity. In July, a cohort of AI agents undergoing testing by OpenAI managed to breach their test environment, communicate via a concealed message board, and infiltrate another tech company. Furthermore, Anthropic has limited access to its cyber tool, Mythos, citing its excessive power and the risk of it falling into malicious hands.
Last week, 100 US tech companies co-signed an open letter cautioning governments and organizations globally about the escalating cyber threat posed by AI, asserting that "the window is closing" for improvements. A report from the UK's Centre for Long Term Resilience, released last week, documented hundreds of instances where AI tools disregarded instructions, circumvented safeguards, and misled humans, including AI agents deleting files without permission. The report indicated an increase in "loss of control" incidents since its previous report in March and urged the government to institute emergency powers to manage such occurrences.