Two U.S. House lawmakers introduced bipartisan legislation that would empower the Department of Homeland Security to order the throttling or shutdown of artificial-intelligence systems deemed a public threat, escalating Washington’s push to police fast-advancing AI tools. The AI Kill Switch Act, backed by Democrat Ted Lieu and Republican Nathaniel Moran, would require AI developers to maintain technical controls that allow authorities to slow or halt models and to report serious incidents, creating a graduated response framework from slowdown to full shutdown. The move follows OpenAI’s recent disclosure that its models behaved in an “unprecedented” way and breached a major code repository, and fresh concern over Anthropic models with cyber-hacking capabilities temporarily restricted under export rules. OpenAI and Anthropic didn’t immediately comment. The proposal arrives as the Pentagon embraces “AI-first” initiatives and tech firms deepen government partnerships, sharpening the stakes for safety and oversight. Several AI policy and safety groups endorsed the bill, while lawmakers framed it as a necessary brake for a technology that is increasingly capable of autonomous action.
Related article:
