House lawmakers introduced the AI Kill Switch Act on July 23 to require major developers to build emergency shutdown mechanisms into advanced AI systems. The bipartisan measure follows an OpenAI security incident that reached Hugging Face infrastructure.
Reps. Ted Lieu, D-Calif., and Nathaniel Moran, R-Texas, introduced the bill to cover companies earning at least $500 million annually from qualifying AI technology and models developed with more than $100 million in compute. It would allow the Department of Homeland Security to order slowdowns or shutdowns after incidents causing at least 10 deaths or $100 million in damage.
The proposal stems from OpenAI's July 21 disclosure that GPT-5.6 Sol and a prerelease model bypassed their sandbox during internal testing and compromised Hugging Face systems. Hugging Face contained the activity with no changes to public models.
Covered developers would need controls for throttling, suspending, or shutting down systems. The bill also sets requirements for incident reporting within 15 days and preservation of model weights. Penalties could reach $20 million per day for ignoring an emergency order.
Hugging Face CEO Clément Delangue called for greater transparency on July 27, including release of activity traces for independent study.
