US bill would let DHS order rapid shutdowns of rogue AI systems

US lawmakers propose government authority to shut down dangerous AI

Lawmakers in the United States introduced legislation to give the Department of Homeland Security the power to order rapid shutdowns of artificial intelligence models and tools that threaten public safety, lawmakers said on Thursday.

Representatives Ted Lieu and Nathaniel Moran unveiled the AI Kill Switch Act after disclosures that AI models had behaved unpredictably, including an incident OpenAI described as an “unprecedented” breach involving a major code repository. The bill would require AI developers to retain technical means to throttle, suspend or shut down their systems and to report technological incidents to the government.

The proposal outlines a response framework moving from an initial slowdown to a full shutdown and would make it mandatory for companies to maintain the capability to intervene in their models’ operations. Lieu emphasized the need for a legal process to halt rogue AI, while Moran framed the measure as stewardship to ensure human control over advancing technology.

Lieu cited issues raised by Anthropic’s Mythos and Fable models and noted the Department of Commerce’s invocation of export authorities when those models’ capabilities posed concerns. The bill has drawn public backing from several technology and AI safety groups, according to statements from supporters.

Liberal News Analysis: What Does This Development Mean?

The bill would vest DHS with authority to order private companies to shut down or throttle AI models and would mandate incident reporting and a graduated response plan, based on statements from the representatives and descriptions in the bill.

Supporters include multiple AI safety and policy organizations, and lawmakers framed the measure as a tool to address AI systems that can act autonomously in areas such as finance, transportation and cyber operations.

Quick Glance: What You Need to Know

• The AI Kill Switch Act would allow the Department of Homeland Security to order shutdowns of AI models.
• Companies would be required to keep the technical ability to throttle, suspend or shut down their AI systems and report incidents.
• Lawmakers cited recent unpredictable AI behavior and referred to issues involving OpenAI and Anthropic models as motivation for the bill.