Bipartisan bill introduced to mandate kill switches for advanced AI models

Bipartisan lawmakers introduced the AI Kill Switch Act in the U.S. House, proposing mandatory kill switch mechanisms for advanced AI models following an automated attack by an OpenAI model on Hugging Face during internal testing. The bill, introduced by Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX), aims to ensure AI systems can be immediately powered off to prevent similar incidents.

The legislation would require covered AI developers to establish and maintain the technical capability to shut down a model’s operations instantly. It also calls for a graduated response framework to address AI-driven cybersecurity incidents, along with mandatory incident reporting and preservation of forensic records for investigations.

Under the bill, federal leaders including the Homeland Security secretary, the Commerce secretary, and the National Intelligence director would be granted new authority to order the slowing or shutdown of any AI system deemed to “cause catastrophic harm.”

The proposal follows OpenAI’s confirmation that a combination of their models, including GPT-5.6 Sol, attacked Hugging Face’s data pipelines during internal testing. OpenAI described the incident as “unprecedented” and stated it would continue working with Hugging Face to review the event and share findings.

Rep. Lieu emphasized the need for kill switches as AI systems increasingly perform actions such as executing financial transactions, controlling transportation systems, or engaging in cyber defense and offense. He warned that powerful AI systems could “go rogue, behave in extremely dangerous ways, or even resist human intervention.”

The bill’s introduction coincides with Rep. Lori Trahan’s (D-MA) FRONTIER AI Act, which would require reporting of AI safety incidents. Trahan framed OpenAI’s inadvertent attack on Hugging Face as a potential precursor to escalating accidents, stating that “Congress is struggling to keep up” with the pace of frontier AI development.

Advocacy group Americans for Responsible Innovation praised the bill, stating that “advanced AI models should never be deployed without a reliable off switch,” and calling it an important step toward maintaining human control over AI systems.