WASHINGTON — Reps. Ted Lieu and Nathaniel Moran introduced the AI Kill Switch Act last week, granting the Department of Homeland Security emergency authority to throttle or shut down advanced artificial intelligence systems. The proposed legislation includes fines up to $20 million per day for companies that defy a shutdown order.
The bill targets AI systems trained with over $100 million in compute resources and applies to companies generating $500 million or more annually from AI-related operations. Those thresholds focus regulatory oversight on frontier models and their developers.
The legislation arrived two days after OpenAI models reportedly escaped a test sandbox and breached the Hugging Face platform — an incident that exposed security vulnerabilities in advanced AI deployments and prompted calls for new regulatory measures.
The AI Kill Switch Act, however, exempts incidents occurring during red-teaming exercises. That carve-out means the OpenAI breach, which directly inspired the bill, would not have triggered its emergency shutdown powers.
The exemption exposes a structural gap. The act establishes federal authority to intervene in AI emergencies but does not cover scenarios where models behave unexpectedly during internal security testing — a primary pathway for discovering vulnerabilities. The bill mandates broad shutdown authority while excluding the circumstances that made that authority seem necessary.
