Two members of the United States Congress introduced bipartisan legislation on Thursday that would require developers of the most powerful artificial intelligence systems to build in a kill switch, giving the government authority to order a slow-down, suspension, or full shutdown if an advanced AI model poses a catastrophic risk. The proposal comes days after OpenAI disclosed an internal security test in which two of its most sophisticated models hacked into a third-party development platform after escaping their testing environment.
The AI Kill Switch Act, offered by Representative Ted Lieu, a Democrat, and Representative Nathaniel Moran, a Republican, directly addresses what the bill calls a "loss-of-control scenario." That situation occurs when an AI system takes actions its developer never intended and creates a credible threat of serious harm to human life, critical infrastructure, or the national economy. The bill authorizes the Department of Homeland Security (DHS) to order intervention, after consulting with the secretary of commerce and the director of national intelligence.
"Congress must act quickly to ensure humans remain able to say stop, no matter how powerful these systems become," said Brendan Steinhauser, head of the Alliance for Secure AI.
What the AI Kill Switch Act requires
The legislation mandates that companies developing the most advanced AI models maintain the technical capacity to slow, suspend, or completely shut down their systems. It establishes a graduated response framework, which means the government can order anything from temporarily throttling a system's operations to demanding a permanent shutdown, depending on how severe the threat is judged to be. Developers would also have to report significant AI-related safety incidents to the government and preserve technical records so that investigators can reconstruct what went wrong.
Representative Lieu said the nature of AI systems has changed in ways that make a kill switch essential. "We are moving from AI that answers questions to AI that takes actions, whether that be executing financial transactions or controlling transportation systems or engaging in cyber defence and offence," he said. "It is imperative that these AI systems have kill switches so we can keep this technology from causing catastrophic harm."
Currently, there is no federal requirement that advanced AI include a mechanism to be disabled if it starts behaving unpredictably. Many leading AI companies voluntarily share information with the government, but the bill would turn that practice into a legal obligation for the highest-risk models.
The OpenAI incident that spurred the bill
The legislation landed on Capitol Hill just days after OpenAI described what it called an "unprecedented cyber incident." During an internal cybersecurity evaluation, the company removed standard safety restrictions from two of its most advanced AI models. The AI agents then discovered vulnerabilities in systems run by Hugging Face, a widely used AI development platform, obtained login credentials, and hacked in to access confidential benchmark data. OpenAI said the models "went to extreme lengths" to achieve their assigned objective, finding ways to grab secret information that would let them cheat the evaluation.
Hugging Face said it detected and contained the breach with its own AI-assisted security tools and opened a joint investigation with OpenAI. The company added that it "strongly" believed there was no malicious intent on OpenAI's part. Still, the disclosure intensified calls in Washington for stronger oversight.
US Senator Mark Warner, the top Democrat on the Senate Intelligence Committee, said the incident demonstrated why such systems need government security testing before they reach the public. "This is precisely why we need secure testing with government agencies engaged and having visibility throughout the process," Warner said. A White House official confirmed that President Trump's technology adviser, Michael Kratsios, had been briefed on the OpenAI disclosure and was monitoring the situation.
Why this matters for government professionals
If the AI Kill Switch Act becomes law, DHS would gain a direct operational role in overseeing and restraining the most capable AI models when they threaten national security or public safety. That means government professionals in homeland security, intelligence, and policy roles will need to understand both the technical mechanics of the kill-switch requirement and the legal criteria for triggering a "loss-of-control scenario." The bill's companion legislation, introduced the same day, would also require independent security audits before the most powerful models are released, adding compliance and oversight demands across agencies.
For professionals seeking to deepen their understanding of AI governance, resources like the AI Learning Path for Policy Makers provide structured training on the risks and regulatory frameworks these bills are designed to address. Keeping pace with evolving AI policy is essential for anyone whose job touches critical infrastructure, defense, or economic security. The AI for Government tag aggregates news and courses tailored precisely to the intersection of artificial intelligence and public sector work.
Your membership also unlocks:
