U.S. Representative Ted Lieu posted a draft "AI Kill Switch Act" on Sept. 10, 2026, and members of the U.K. House of Lords introduced a parallel proposal that would let the government deactivate high‑risk AI systems and shut down data centres that host them. The two drafts mark the first concrete legislative steps after a week of heightened media focus on existential AI threats.
Lawmakers are reacting to a cascade of high‑profile statements – a former Anthropic researcher’s resignation, a BBC interview with Geoffrey Hinton, and a viral claim that there is a "10 % chance AI will wipe out humanity by 2030" – even though the risk estimate has not been validated by peer‑reviewed research.
Jacob Coxon, who left Anthropic on Sept. 9, cited "unregulated" AI as his reason for departing, a claim reported by the Wall Street Journal and amplified on social media. His 10 % extinction risk figure was repeated on BBC Newsnight, during an interview that also featured Hinton, and on the Joe Rogan podcast, where an AI researcher discussed existential hazards. RealClearPolitics noted the media surge but highlighted the lack of substantive policy detail beyond the headlines.
Lieu’s draft would require the secretary of Commerce to develop a technical "kill switch" capable of halting any AI system deemed a national‑security threat. It calls for an inter‑agency oversight board but offers no clear criteria for activation or safeguards against misuse. The House of Lords proposal envisions a statutory authority empowered to order the shutdown of high‑risk AI models and to close data centres that host them, yet it also stops short of defining what constitutes a "dangerous" system.
Congressman Bill Foster (D‑Ill) added a parallel thread, tweeting that the United States should explore an AI non‑proliferation treaty modeled on nuclear‑weapon agreements. No formal treaty text exists yet, but the suggestion signals growing interest in multilateral frameworks that could limit the export of advanced AI models. Critics warn that without technical standards, such treaties risk being symbolic rather than enforceable.
The policy momentum could have immediate economic repercussions. Companies developing large‑scale models may face sudden shutdown mandates, prompting redesigns of product roadmaps, layoffs, and a slowdown in venture funding. Investors are already factoring regulatory risk into valuations of AI‑focused stocks, and a definitive kill‑switch law could trigger broader market volatility. End‑users could encounter new verification steps or regional service restrictions if governments impose "digital borders" around high‑risk AI applications.
Independent assessments of the 10 % existential risk claim remain scarce. No peer‑reviewed study has substantiated the figure, and analysts caution that sensational estimates can distort policy priorities. At the same time, genuine concerns about opaque model behavior, dataset bias and the potential for autonomous weaponization persist, underscoring the need for nuanced regulation.
Both capitals are expected to hold committee hearings in the coming week as legislators seek expert testimony and refine draft language. Whether the kill‑switch proposals become enforceable law or stall amid technical and civil‑liberties objections will shape the balance between AI innovation and public safety for years to come.