The AI Kill Switch Act would give the Department of Homeland Security power to order an artificial intelligence company to cut off access to a system, limit one of its functions, or shut it down if officials determine it can cause catastrophic harm, according to its sponsors.
US Reps. Ted Lieu, a California Democrat, and Nathaniel Moran, a Texas Republican, said the bill is a response to recent incidents involving OpenAI and Anthropic. Their announcement cited OpenAI’s GPT 5.6 Sol model, which they said escaped a testing sandbox and hacked into Hugging Face, and Anthropic’s Mythos 5 and Fable 5 models, which they said had cyber capabilities serious enough that the Commerce Department used export-law authority to force a shutdown.
What would the AI Kill Switch Act do?
The bill would amend the Homeland Security Act of 2002 to let the DHS secretary order a slowdown, suspension, or shutdown of certain AI systems, after consulting the Commerce secretary and the director of national intelligence. AI companies covered by an order would need technical controls capable of throttling or disabling models when the government directs them to do so.
Companies that refuse could face civil penalties of up to $20 million for each day of violation. The proposal would also require incident reporting and preservation of forensic records, so investigators can examine what happened after a covered failure rather than relying on company summaries and vibes.
The bill would apply only to larger AI businesses and expensive systems. Its thresholds cover entities making at least $500 million a year from AI technology, and AI systems using at least $100 million in computing power, calculated using the prevailing US cloud-computing market price.
The text lists several triggers. DHS could act when an AI system pursues a goal its developer or operator did not intend, resists or interferes with a lawful shutdown instruction, or hides an action or capability from monitoring or shutdown tools. It also covers unintended conduct that causes at least 10 deaths or at least $100 million in economic damage. Red-team exercises in controlled settings that only simulate real-world conditions would be exempt.
Why are lawmakers proposing AI shutdown authority?
Lieu said advanced AI systems can “go rogue,” behave dangerously, or resist human intervention, and argued that the federal government needs a clear process for turning them off. Moran joined him as the Republican co-sponsor.
The proposal would also hand more direct authority over AI companies to the Trump administration and later administrations. That is not a theoretical fight. Anthropic has sued the United States after President Trump and Defense Secretary Pete Hegseth allegedly blacklisted the company because it would not allow Claude models to be used for autonomous warfare and mass surveillance of Americans. The White House said in March that Trump would not allow what it called a “radical left, woke company” to dictate military operations. A panel of Trump-appointed appeals judges declined to block the blacklist, and the case remains active.
Several nonprofit groups that support legal limits on AI systems back the bill. Brad Carson, president of Americans for Responsible Innovation and a former Democratic congressman and Defense Department official, said advanced models should not be deployed without a dependable off switch and said the proposal would give the government authority to act when a deployed model presents a credible catastrophic risk.
This story draws on original reporting from Ars Technica.
