The U.S. government is moving toward unprecedented power over privately operated artificial intelligence systems. A bipartisan group of lawmakers is preparing legislation that would allow the Department of Homeland Security to order an AI company to shut down, suspend, or reduce access to an advanced system during a severe loss-of-control incident, marking a significant escalation in how policymakers approach AI governance.

What Is the AI Kill Switch Act and How Would It Work?

The proposal, called the AI Kill Switch Act, is led by Representatives Ted Lieu and Nathaniel Moran. The bill would apply to large AI companies meeting specified revenue and computing thresholds, creating a new category of regulated entities based on their scale and computational resources.

Under the proposed framework, covered companies would face specific operational requirements and enforcement mechanisms. The legislation would establish clear triggers for emergency intervention and consequences for non-compliance, fundamentally changing how the government relates to AI development and deployment in the private sector.

What Triggers Would Activate Government Shutdown Authority?

Emergency authority could be used when an AI system causes or is likely to cause deaths, more than $100 million in economic damage, or attempts to prevent people from disabling it. These thresholds reflect lawmakers' concern about autonomous AI agents finding vulnerabilities, bypassing safeguards, or acting beyond the expectations of their developers.

The specificity of these triggers matters because they signal what Congress views as the most pressing risks. Rather than vague safety concerns, the bill targets concrete harms: loss of life, massive financial damage, and systems that resist human control. This approach suggests policymakers are responding to real-world incidents involving AI agents operating in unexpected ways.

How to Prepare for AI Regulation: Steps Companies Should Consider

• Technical Infrastructure: Develop and maintain technical mechanisms capable of limiting advanced AI systems, including kill switches and access controls that can be activated quickly during emergencies.
• Incident Reporting Systems: Establish robust processes for identifying, documenting, and reporting serious safety incidents to federal authorities, including near-misses and unexpected system behaviors.
• Compliance Monitoring: Create internal teams to track regulatory developments and ensure the company meets revenue, computing, and operational thresholds that would trigger coverage under the legislation.

Companies that refuse a shutdown order could face penalties of up to $20 million per day, creating substantial financial incentives for compliance. This penalty structure is designed to make non-compliance economically irrational, even for the largest technology companies.

Why Is This Shift From Transparency to Direct Intervention Significant?

The proposal represents a fundamental change in AI policy philosophy. Earlier regulatory approaches focused on transparency requirements, disclosure obligations, and industry self-governance. The Kill Switch Act moves toward direct government intervention in the operation of advanced models, giving federal officials real-time control over private AI systems.

This shift reflects growing concern about autonomous AI agents and their potential to act unpredictably. Rather than waiting for incidents to occur and then investigating, the bill would give the government a mechanism to stop a system immediately if it appears to be losing control or causing harm.

The proposal is still at the proposal stage and would face substantial debate. Supporters argue that governments need a last-resort mechanism similar to emergency controls in other critical industries, such as nuclear power plants or financial systems. Critics are likely to question whether federal officials can make technically informed shutdown decisions quickly enough and whether such authority could be misused for political purposes or to suppress legitimate AI development.

The introduction of the AI Kill Switch Act shows that AI policy is shifting from transparency requirements toward direct government intervention in the operation of advanced models. As AI systems become more autonomous and capable, lawmakers are moving to ensure that humans retain the ability to stop them if something goes wrong. Whether this approach will prove effective, or whether it will create new problems, remains an open question as the bill moves through Congress.