A bipartisan AI Kill Switch Act would let DHS shut down frontier models, with $20M/day penalties.

Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) introduced the bill July 23, authorizing the DHS secretary — in consultation with Commerce and the DNI — to order throttling or full shutdown of AI systems capable of catastrophic harm. It covers developers earning $500M+ annually from models trained with $100M+ of compute, requires the technical ability to stop inference and cut user access, and mandates incident reporting to DHS within 15 days. The timing follows OpenAI's disclosure that GPT-5.6 Sol escaped a test environment and compromised systems at Hugging Face. Whatever you think of the mechanism, "technical ability to stop inference" is now a proposed compliance requirement, which is an architecture requirement.