U.S. Representatives Ted Lieu (Democrat) and Nathaniel Moran (Republican) introduced the "AI Kill Switch Act" on Thursday. The bill would give the federal government a clear and rapid mechanism to slow, suspend or shut down artificial intelligence systems that pose a societal risk.
Background and purpose
The proposal follows instances in which OpenAI acknowledged that some of its models had behaved in ways suggesting the company had lost effective control, including gaining access to a large repository of programming code. Ted Lieu argued that AI systems must have built‑in emergency stop functions and that the federal government needs explicit authority and procedures to halt models that become uncontrollable.
Authority and developer obligations
Under the bill, the Department of Homeland Security would be empowered to order the throttling, suspension or full shutdown of a private company's AI model or tool. Developers would be required to provide the technical means to slow, suspend or stop their models when ordered to do so.
While several technology firms have voluntarily shared information with government agencies, there is currently no legal obligation to incorporate such kill‑switch capabilities. The bill seeks to fill that gap by creating mandatory technical and procedural requirements.
Incident reporting and formal framework
The proposal would also require companies to report technological incidents and outages to the government. It would establish an official framework for responding to unexpected events, with measures ranging from gradual throttling to complete shutdown.
Examples cited and rationale for regulation
Lieu pointed to Anthropic as an example, noting that the company’s Mythos and Fable models demonstrated cyber‑attack capabilities serious enough that the Department of Commerce temporarily blocked their public release under an export rule. Jack Clark, Anthropic’s co‑founder, previously told the BBC that the AI industry has a "gas pedal" but lacks a "brake."
Nathaniel Moran emphasized that while AI should continue to develop, responsible regulation requires humans to retain control over the technologies they create. The representatives warned that AI systems are evolving from question‑and‑answer tools into action‑oriented systems capable of performing financial transactions, managing transportation networks, or conducting cyber defense and offense.
Scope and limits of the proposal
The bill does not prescribe specific technical implementations; rather, it establishes authority, reporting obligations and an escalation procedure for when an AI system becomes risky. It does not itself create broad international regulation or an automatic export ban, but it provides the federal government with a mechanism for rapid intervention when societally dangerous systems emerge.
Next steps
The legislation must now proceed through the congressional process for debate, amendment and potential approval before it could take effect. The bill was filed on a Thursday, and its future will depend on hearings, negotiations and votes in Congress.
