Two senior lawmakers are preparing federal legislation that would grant the Department of Homeland Security authority to force AI companies to shut down or significantly throttle their systems in response to emergencies.
Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) plan to introduce the "AI Kill Switch Act" this week. The bill reflects growing congressional concern about AI systems operating without emergency shutdown mechanisms, particularly as models grow more capable and autonomous.
The legislation targets a real operational gap. Most major AI systems lack standardized emergency protocols that government agencies can activate. The DHS, which oversees national cybersecurity and critical infrastructure, would gain explicit power to order AI systems offline or degraded if regulators determine they pose imminent threats to national security, public safety, or critical infrastructure.
The bipartisan backing signals broad concern about AI safety oversight. Lieu has consistently pushed for stronger AI regulation, while Moran's involvement reflects Republican support for this particular intervention. Neither party views emergency shutdown authority as ideologically problematic, though the specifics matter enormously.
Key questions remain unanswered. The bill's text, not yet public, will determine enforcement mechanisms, response timelines, and whether companies face penalties for non-compliance. It will also clarify what "throttling" means technically and how DHS determines when a shutdown is warranted versus merely precautionary.
Industry response will likely emphasize operational disruption risks. AI companies may argue that emergency powers could be abused or deployed prematurely, disrupting legitimate services. They may also raise questions about whether DHS has the technical expertise to make real-time assessments about which systems need shutdown.
The proposal addresses a legitimate gap in AI governance. Federal agencies responsible for disaster response, cybersecurity, and critical infrastructure protection currently lack direct intervention tools. As AI systems integrate deeper into banking, power grids, and communications networks