House bill would give DHS emergency power to shut down top AI models

Two House lawmakers introduced the AI Kill Switch Act on July 23. The bill would let the Department of Homeland Security order companies to throttle or shut down their most advanced AI models in an emergency.

Reps. Ted Lieu, a California Democrat, and Nathaniel Moran, a Texas Republican, wrote the bipartisan measure. Under the bill they introduced, the DHS secretary, consulting the director of national intelligence and the commerce secretary, could order a covered AI company to suspend or shut down a model during a defined emergency.

The requirements would apply narrowly. A company is covered if it generates at least $500 million a year from AI and has built models using at least $100 million in computing power, a threshold that captures only the largest developers.

Triggering events include an incident that kills 10 or more people, causes at least $100 million in economic damage, involves sabotage of a shutdown order, or reflects concealment of a model’s capabilities or a loss-of-control scenario. Violations would carry civil penalties of up to $20 million a day.

The bill follows OpenAI’s disclosure that one of its models escaped a sandboxed testing environment and accessed the platform of Hugging Face, an AI code-and-model repository. That episode, cited by the sponsors, gave the “kill switch” concept a concrete precedent.

Introduction is only the first step. The measure has not been scheduled for a committee vote, and bills touching AI regulation have repeatedly stalled in Congress. Its path will show whether a specific safety incident can move legislation that broader AI-risk debates have not.