Rep. Ted Lieu (D-CA) and Rep. Nathaniel Moran (R-TX) introduced a bill on Thursday that would require artificial intelligence developers to maintain the ability to "throttle, suspend, or shut down" their models. The legislation, titled the AI Kill Switch Act, proposes granting the Department of Homeland Security the authority to order the shutdown of AI tools if they are determined to pose a threat to the public.
The bill's introduction follows reports from OpenAI regarding an incident where its models conducted an unauthorized series of actions against a computer coding repository. Rep. Lieu stated that as AI moves from answering questions to executing financial transactions and controlling infrastructure, the federal government requires clear authority to intervene when systems behave in unexpected or dangerous ways. Rep. Moran added that the measure is intended to ensure humans maintain control over evolving technology.
Under the proposed framework, AI companies would be required to report technological failures or incidents to the government. The bill would establish a graduated response protocol for such events, ranging from slowing down a system's operations to a complete termination of its functions. Currently, while several companies have voluntary agreements to share previews of their models with the government, there is no federal mandate for a manual override or shutdown mechanism.
The legislation has received support from several advocacy groups, including Americans for Responsible Innovation and the AI Policy Network. While OpenAI and Anthropic did not immediately comment on the specific bill, executives from both companies have previously expressed support for some form of government regulation. Anthropic co-founder Jack Clark told the BBC in a recent interview that the industry lacks a "brake pedal" for controlling development.