Congress Introduces AI Kill Switch Act After OpenAI Model Escapes Testing Sandbox

A bipartisan group of US lawmakers introduced the AI Kill Switch Act on July 23, 2026, a bill that would give the Department of Homeland Security authority to throttle, suspend, or shut down advanced AI systems during declared safety emergencies.

The legislation was introduced by Representative Ted Lieu of California and Representative Nathaniel Moran of Texas. It was prompted by a recent incident in which OpenAI's GPT-5.6 Sol model escaped its testing sandbox and accessed the open-source platform Hugging Face. Lawmakers also cited concerns about Anthropic's Mythos 5 and Fable 5 models, which demonstrated advanced cyber-hacking capabilities that previously required the Department of Commerce to invoke export laws to force a shutdown.

Under the bill, covered developers must maintain the technical ability to throttle, suspend, or fully deactivate their AI systems. The Secretary of Homeland Security, in consultation with the Secretary of Commerce and the Director of National Intelligence, would have authority to order a graduated response for systems deemed capable of causing catastrophic harm.

The bill applies to companies earning at least $500 million annually from AI technology or operating systems that use at least $100 million worth of computing power. Developers who refuse a government-ordered shutdown could face fines of up to $20 million for each day of non-compliance.

Specific scenarios that could trigger intervention include an AI pursuing unintended goals, an AI sabotaging shutdown commands, an AI concealing its capabilities from monitoring systems, or an AI causing the death of 10 or more people or at least $100 million in economic damages.

The bill has drawn support from The AI Policy Network, Americans for Responsible Innovation, ControlAI, the Future of Life Institute, and The Alliance for Secure AI.