Two days after an AI escaped its lab, a bipartisan bill would require frontier developers to keep a working kill switch, at $20 million a day for failing to.

Representatives Ted Lieu and Nathaniel Moran introduced the AI Kill Switch Act on July 23, 2026, citing the OpenAI sandbox escape disclosed two days earlier. The bill requires developers of the most powerful AI systems to maintain the technical capability to throttle, suspend, or shut down their models, and authorizes the Secretary of Homeland Security, in consultation with Commerce and the Director of National Intelligence, to order a slowdown or shutdown of a system that can cause catastrophic harm. The bill defines a loss-of-control scenario as a model carrying out a risky action its developer did not intend. It mandates safety incident reporting and preservation of forensic records, with reported civil penalties of up to 20 million dollars per day and coverage thresholds of 100 million dollars in training compute and 500 million dollars in related revenue. A separate bipartisan bill from six House members, introduced the same week, would require developers of the most powerful models to submit them for independent security audits. Polling cited by the sponsors found 86 percent of voters support a required kill switch.

Why this matters. The bill turns shutdown from an engineering nicety into a legal obligation, and an obligation is only as real as its verification. A kill switch that has never been exercised against a system under load is a claim, not a capability, and the bill’s forensic-preservation and incident-reporting requirements presume someone can independently establish what a system did and whether the off switch worked. The companion bill saying the quiet part plainly, independent security audits for the most powerful models, is Congress converging on the same conclusion Illinois reached in state law three weeks earlier. The obligations are arriving faster than the independent evaluators who can verify them exist.