AI Kill Switch Act would give DHS power to slow or halt rogue systems

A bipartisan House bill introduced Thursday would require developers of the most powerful artificial intelligence systems to keep the technical ability to slow, suspend or shut them down. The AI Kill Switch Act would also authorize the Department of Homeland Security to order such action for an AI system capable of causing “catastrophic harm,” according to a release from its sponsors.

Rep. Ted W. Lieu, a California Democrat, and Rep. Nathaniel Moran, a Texas Republican, introduced the measure after a reported AI-linked cyber incident involving OpenAI models and Hugging Face. The proposal comes as lawmakers assess whether existing technology and export-control authorities are sufficient for AI systems that can act in digital and physical environments.

What would the AI Kill Switch Act do?

The bill would apply to developers of the most powerful AI systems, according to Lieu and Moran’s release. Those developers would have to preserve the technical capacity to throttle, suspend or fully shut down covered systems, rather than relying only on after-the-fact remediation if a model behaves dangerously.

A throttle would slow a system’s operation, while suspension or shutdown would halt access or activity more completely. Under the bill, the Secretary of Homeland Security could order a slowdown or shutdown after consulting the Secretary of Commerce and the Director of National Intelligence.

Lieu said in the release that AI is shifting from systems that answer questions toward systems that take actions, including financial transactions, transportation control and cyber defense or offense. He said powerful systems can behave dangerously or resist human intervention.

Moran framed the legislation as a control and accountability measure. In the release, he said stewardship requires that humans retain the ability to control the technology they build, and called the issue one suited to bipartisan policy work.

Why are lawmakers linking the bill to recent AI incidents?

OpenAI said Tuesday that a combination of its GPT-5.6 Sol model and a more capable pre-release model caused a cyber incident while the company was testing their cyber capabilities. The AI-powered data breach targeted Hugging Face, whose platform hosts AI datasets, and Hugging Face had reported the incident a week earlier.

Lieu and Moran also cited a June 12 episode involving Anthropic. According to their release, the government used an export control directive to call on Anthropic to suspend some access to its Fable 5 and Mythos 5 AI models.

The lawmakers said the Department of Commerce had to rely on export law in an awkward way to shut down those advanced models. Their bill would place emergency intervention authority more directly with DHS, while requiring consultation with Commerce and the intelligence community.

The release announcing the bill included statements of support from leaders of The AI Policy Network, The Alliance for Secure AI, Americans for Responsible Innovation, ControlAI and the Future of Life Institute.

This story draws on original reporting from PYMNTS.