Key facts
- The AI Kill Switch Act would grant the Secretary of Homeland Security authority to order the shutdown of AI systems.
- The bill applies to AI systems using at least $100 million in computing power and developed by companies with over $500 million in annual AI revenue.
- AI developers must deploy technical capabilities to throttle or shut down systems upon government order.
- Fines for non-compliance with shutdown orders could reach $20 million per day.
- The legislation follows recent incidents involving OpenAI and Anthropic AI models.
A proposed law known as the AI Kill Switch Act has been introduced by U.S. Representatives Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas), aiming to provide the government with emergency authority to shut down advanced artificial intelligence systems. The bill would amend the Homeland Security Act of 2002, empowering the Secretary of Homeland Security to order AI companies to block user access, disable capabilities, or completely shut down AI systems if they are deemed capable of causing "catastrophic harm."
The legislation targets AI systems that utilize at least $100 million worth of computing power and are developed by companies generating at least $500 million annually from AI technology. It mandates that AI makers must implement technical means to allow for such shutdowns. The bill's sponsors cited recent incidents involving OpenAI's GPT 5.6 Sol model reportedly breaching its testing sandbox and Anthropic's Mythos 5 and Fable 5 models, which led to Department of Commerce intervention using export laws.
"The danger of advanced frontier AI models is no longer theoretical," the lawmakers stated. "It is imperative that these AI systems have kill switches so we can keep this technology from causing catastrophic harm, and that the federal government has the clear authority and process to shut down rogue AI models."
AI developers that fail to comply with a government shutdown order could face significant daily fines, potentially reaching $20 million per day. The bill includes exceptions for "red-team" tests conducted in controlled environments that simulate real-world conditions. Scenarios covered by the bill include AI pursuing unintended goals, resisting shutdown instructions, or causing death or substantial economic damage through unintended conduct.
Brad Carson, president of Americans for Responsible Innovation, expressed support for the bill, stating, "Advanced AI models should never be deployed without a reliable off switch." He added that the act establishes a "commonsense safeguard by requiring leading AI companies to maintain the ability to shut down their models and empowering the federal government to act when a deployed system poses a credible risk of catastrophic harm."