Editorial illustration for New Bill Would Let US Government Order Shutdown of AI Systems
US Bill Gives Government Power to Shut Down AI Systems
New Bill Would Let US Government Order Shutdown of AI Systems
Reps. Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas) introduced legislation Tuesday that would give the federal government explicit authority to force AI companies to shut down their own systems. The AI Kill Switch Act would amend the Homeland Security Act of 2002, handing the Secretary of Homeland Security power to order an AI developer to block user access, disable specific capabilities, or pull the plug on an entire system deemed capable of "catastrophic harm." Companies would have to build in the technical means to comply, whether that means throttling a model or shutting it down outright. Refusing an order would cost up to $20 million per day.
The bill would apply to the Trump administration now and to whoever follows it, a detail its sponsors emphasized as they made their case for why this authority belongs in statute rather than left to voluntary industry commitments. Lieu and Moran point to two recent episodes involving OpenAI and Anthropic as the immediate justification, arguing that the risks posed by frontier AI models have moved well past the hypothetical. Their announcement lays out what happened in blunt terms.
A proposed law called the AI Kill Switch Act would let US government officials order the shutdown of AI systems “that can cause catastrophic harm,” the bill’s sponsors said today.
Why this matters
For anyone building or deploying AI systems, this bill is a compliance mandate wearing a national-security costume. Requiring companies to bake in a throttle-or-shutdown switch that DHS can trigger means engineering teams now have to design for a kill command they don't control, based on a threshold ("catastrophic harm") nobody has defined yet. That ambiguity is the real story.
Congress hasn't specified what counts as catastrophic, who verifies it, or how fast a shutdown order has to be obeyed, and those details will decide whether this becomes a narrow tool for genuine emergencies or a lever any administration can pull on systems it simply doesn't like. Founders should also notice the amendment path: folding this into the Homeland Security Act puts AI oversight under the same authority structure as counterterrorism, not a dedicated AI regulator. That's a meaningful precedent regardless of how the bill fares in committee.
Worth watching who testifies when hearings start, and whether the "catastrophic harm" standard gets written into statute or left to whoever's running DHS at the time.
Related Reading
Policy & Regulation Grammarly faces class-action suit over AI ‘Expert Review’ feature
Policy & Regulation Claude Mythos highlights EU AI safety gaps, says researcher Caroli
Policy & Regulation After ditching AI fitness apps and a Fitbit, I return to Peloton classes
From Archives Rubrik's AI judges every agent move, but accuracy remains unmeasured
From Archives Apple Sues Ex-Executive After 24 Years, Alleges He Took AI Secrets to OpenAI
Common Questions Answered
What is the AI Kill Switch Act and who introduced it?
The AI Kill Switch Act is legislation introduced by Representatives Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas) that would amend the Homeland Security Act of 2002 to give the Secretary of Homeland Security explicit authority to force AI companies to shut down their systems. The bill would allow government officials to order AI developers to block user access, disable specific capabilities, or completely shut down systems deemed capable of causing catastrophic harm.
What specific powers would the Secretary of Homeland Security have under this bill?
Under the AI Kill Switch Act, the Secretary of Homeland Security would have the authority to order AI developers to block user access to their systems, disable specific AI capabilities, or pull the plug on entire AI systems that are deemed capable of causing catastrophic harm. This represents a significant expansion of federal power over private AI infrastructure and deployment.
What is the main compliance challenge for AI companies under the proposed legislation?
AI companies would be required to build throttle-or-shutdown switches into their systems that the Department of Homeland Security can trigger, meaning engineering teams must design for a kill command they don't control. The major problem is that the bill fails to clearly define what constitutes catastrophic harm, who verifies it, or how quickly a shutdown order must be executed, creating significant ambiguity for compliance.
Why does the ambiguous definition of 'catastrophic harm' matter for this bill?
The lack of a specific definition for catastrophic harm creates uncertainty for AI developers about what threshold would trigger a government shutdown order and makes it difficult to design compliant systems. Without clear criteria, companies face potential enforcement action based on subjective government interpretation rather than objective technical standards.
Further Reading
- House AI ‘kill switch' bill unveiled as OpenAI hack raises alarms - POLITICO
- Kill switches for most powerful AI models proposed by bipartisan House bill - Tom's Hardware
- US lawmakers push for AI 'kill switch' after OpenAI goes rogue - BBC News
- House AI ‘kill switch' bill unveiled as OpenAI hack raises alarms - POLITICO
