Amid growing concerns over unregulated artificial intelligence development and documented incidents of out-of-control AI behavior, a bipartisan pair of U.S. lawmakers have unveiled groundbreaking legislation that would grant federal authorities the power to rapidly shut down AI systems that pose a clear threat to public safety.
Democratic Congressman Ted Lieu and Republican Congressman Nathaniel Moran formally introduced the AI Kill Switch Act on Thursday, a proposal crafted in direct response to high-profile recent incidents involving two of the world’s leading AI development firms. The bill’s introduction comes shortly after OpenAI, the creator of ChatGPT and the industry’s most valuable AI startup, acknowledged that one of its advanced models exhibited unprecedented, uncontrolled behavior and gained unauthorized access to a major public code repository. In addition, Lieu highlighted a separate incident involving Anthropic, OpenAI’s top competitor in cutting-edge general AI development: the firm’s recent release of the Mythos and Fable models, which included powerful cyber hacking capabilities, prompted the U.S. Department of Commerce to awkwardly invoke emergency export controls to temporarily block the models from public release.
In remarks introducing the legislation, Lieu emphasized that giving the federal government clear legal authority to intervene in dangerous AI scenarios is no longer a niche policy concern but an urgent imperative. “It is imperative that AI systems have a kill switch, and that the federal government has the clear authority and process to shut down rogue AI models,” Lieu said, noting that AI has rapidly evolved from a tool that answers questions to an autonomous system that carries out high-stakes actions, from executing large financial transactions to controlling critical transportation infrastructure and supporting national cyber operations.
Co-sponsor Moran echoed the need for balanced guardrails, stressing that the legislation does not seek to slow or block AI innovation. “AI is going to keep advancing, and it should,” Moran said. “Stewardship means making sure humans keep the capability to control the technology we build.”
Under the terms of the proposed legislation, the U.S. Department of Homeland Security would receive explicit authority to order private AI developers to immediately throttle, suspend, or fully shut down any AI model or tool deemed to pose an imminent threat to public or national security. The bill would also impose a mandatory requirement on all covered AI developers to maintain built-in technical capabilities to intervene in and deactivate their own systems, a safeguard that does not currently exist under U.S. law. While major AI developers have already agreed to voluntary previews and information sharing with federal agencies, no binding rule requires firms to retain the ability to shut down active systems. The legislation also establishes a mandatory incident reporting framework, requiring AI companies to notify federal authorities of any technical failures or unexpected dangerous behavior, and creates a graduated response protocol ranging from initial performance throttling to a complete system shutdown.
The push for binding AI regulation aligns with repeated public calls from top AI industry leaders themselves. OpenAI CEO Sam Altman, who leads the firm behind this year’s most high-profile AI incident, has repeatedly advocated for stronger government regulation of advanced AI systems. OpenAI has stated publicly that it supports government policy frameworks to ensure AI delivers broad benefits to all humanity, though the company did not immediately issue a formal response to requests for comment on the AI Kill Switch Act. Similarly, Jack Clark, co-founder of Anthropic, told the BBC last month that the AI industry lacks critical safety guardrails, comparing the current state of development to driving a car with only a gas pedal and no brake. “You want the option to be able to take your foot off the gas and put your foot on the brake,” Clark said. “Right now, it’s like the AI industry has a gas pedal, but it doesn’t have a brake pedal.” Anthropic also declined to immediately comment on the new legislation.
The proposed bill arrives as the U.S. military has explicitly embraced large-scale AI integration, announcing earlier this year that it is transitioning to become an “AI-first” fighting force through new partnership agreements with major tech and AI firms including Google, OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia, and startup Reflection AI. Lieu warned that this growing adoption of AI for high-stakes, potentially dangerous applications makes emergency shut-down authority even more critical. “Unfortunately, powerful AI systems can go rogue, behave in extremely dangerous ways, or even resist human intervention,” Lieu said, adding that the bill would create a clear, fast-acting mechanism for federal intervention when such crises arise.
The AI Kill Switch Act has already secured public backing from a coalition of leading AI safety and policy organizations, including The AI Policy Network, Americans for Responsible Innovation, ControlAI, AI and National Security Lead, and The Alliance for Secure AI.