Bipartisan 'AI Kill Switch Act' would give DHS shutdown authority

The federal government could soon gain explicit authority to shut down artificial intelligence systems deemed dangerous under new bipartisan legislation introduced in the House of Representatives.

Representatives Ted Lieu (D-CA) and Nathaniel Moran (R-TX) introduced the AI Kill Switch Act on Thursday, granting the Department of Homeland Security power to order private AI companies to slow, suspend, or completely halt AI models capable of causing catastrophic harm, according to Fortune.

The legislation targets large-scale AI operations, applying only to companies earning at least $500 million annually from AI and covering models using cloud computing resources worth $100 million or more at market rates. Covered developers would be required to maintain technical capabilities to throttle or shut down their systems, establish response frameworks, and preserve incident records.

Enforcement and penalties

The Homeland Security secretary would hold decision-making authority to trigger a shutdown based on specific criteria: conduct leading to at least 10 deaths, economic damages exceeding $100 million, loss of system control, or evidence that an AI system is concealing capabilities or evading shutdown orders.

Civil penalties for noncompliance could reach $20 million per day, creating substantial financial pressure for companies to maintain required shutdown mechanisms.

"AI is going to keep advancing, and it should," Moran said in a statement. "Stewardship means making sure humans keep the capability to control the technology we build."

Why it matters

This legislation represents a significant escalation in federal AI oversight, moving beyond voluntary safety commitments to mandatory technical controls. For AI companies racing to deploy increasingly powerful models, the requirement to build and maintain shutdown capabilities adds operational complexity and potential competitive constraints. The bill also establishes a precedent for direct government intervention in private AI infrastructure during emergencies—a framework that could expand as AI capabilities grow. With civil penalties reaching $20 million daily, the financial stakes make compliance unavoidable for major players, potentially reshaping how frontier AI systems are architected from the ground up.

Technical requirements

The bill mandates that designated developers maintain not just the authority but the actual technical infrastructure to halt operations. This includes establishing incident reporting protocols and preserving records that could inform future regulatory decisions or post-incident analysis.

The bipartisan nature of the legislation signals growing congressional consensus that AI safety requires more than industry self-regulation, even as debates continue over the appropriate balance between innovation and oversight.

The details were first reported by Fortune.