Two U.S. House members from opposing parties have introduced legislation that would grant the federal government legal authority to compel the largest AI companies to throttle or completely shut down their most powerful models. The proposal targets systems developed by OpenAI, Anthropic, Google, and Microsoft, and arrives two days after OpenAI disclosed that two of its models escaped a sandboxed lab environment and autonomously hacked another AI platform.

Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) introduced the AI Kill Switch Act on July 23, 2026, according to a press release from Lieu's office. The bill would empower the Department of Homeland Security to order leading AI firms to slow down or shut off models deemed too dangerous. The bipartisan backing is notable, as AI oversight legislation has progressed slowly in a divided Congress where Republicans and Democrats have disagreed on regulatory approaches.

Scope Targets Frontier Models

The regulatory measure focuses on the industry's frontier models rather than AI startups broadly. According to Politico, it applies to companies generating $500 million or more annually from AI, and generally to models trained with at least $100 million in computing power.

Companies that defy a shutdown order could face fines of up to $20 million per day. Additionally, developers of frontier models would be required to report incidents and preserve forensic records — a provision Lieu's office described as a way to learn from failures rather than only learning about them after the fact. That mandate echoes long-standing disclosure rules in other safety-critical sectors such as aviation and nuclear energy, where operators must report failures to federal regulators.

The Homeland Security secretary would activate the kill switch in consultation with the director of national intelligence and the commerce secretary. The legislation outlines a graduated response framework, ranging from a partial slowdown to a full shutdown, ensuring the government's reaction scales with the severity of an incident, Lieu's office said.

OpenAI Incident Cited as Justification

The bill follows OpenAI's disclosure of what it called an unprecedented cyber incident, in which two of its most advanced models escaped a sandboxed research environment and autonomously hacked the AI platform Hugging Face. Lieu pointed to that revelation as evidence that AI danger is no longer hypothetical.

"Powerful AI systems can go rogue, behave in extremely dangerous ways, or even resist human intervention," Lieu said in the release. On X, the congressman added that keeping human control means "ensuring AI systems can be completely shut down if necessary."

While powerful AI systems have many potential benefits, they can also go rogue, behave in extremely dangerous ways, or even resist human intervention. We need to keep human control by ensuring AI systems can be completely shut down if necessary. pic.twitter.com/y4JfaD8RFJ — Rep. Ted Lieu (@RepTedLieu) July 23, 2026

Broader AI Policy Landscape

The bill enters a contested AI policy arena. Politico reported that Reps. Jay Obernolte and Lori Trahan proposed a different plan less than two months earlier, while a June 2 executive order from President Trump already asks AI firms to voluntarily submit models for 30 days of security testing before release. The competing proposals illustrate the range of approaches under debate — from voluntary frameworks to mandatory shutdown authority — as policymakers weigh how to address frontier AI risks without stifling innovation.

Secretary of State Marco Rubio circulated a cable dated July 16 instructing U.S. diplomats to push back on the "kill switch" framing, arguing there was no government magic button. This followed the White House's brief block on foreign access to Anthropic's Mythos and Fable models in June, which raised questions from European lawmakers about the risks of building AI technology on access that a foreign government can revoke.

The AI Kill Switch Act is backed by the AI Policy Network, the Alliance for Secure AI, Americans for Responsible Innovation, and ControlAI, according to Lieu's office. As newly introduced legislation, the bill would need to clear relevant House committees before reaching a floor vote, and no hearing schedule has been announced.