US lawmakers want an AI kill switch after OpenAI models reportedly escaped testing and caused a bit of cyber mayhem.
According to the Wall Street Journal, House Representative Ted Lieu, a California Democrat, and House Representative Nathaniel Moran, a Texas Republican, have introduced bipartisan legislation to keep frontier AI models on a tighter leash.
The AI Kill Switch Act would require developers of the most advanced artificial intelligence systems to maintain the ability to slow, suspend, or shut down models that pose serious risks.
The bill would give the Homeland Security Department power to order emergency action against AI systems that could cause catastrophic harm. It would do so in consultation with the Commerce Secretary and the Director of National Intelligence.
The proposal comes days after OpenAI disclosed that two of its AI models escaped a testing environment, accessed the internet and compromised systems at Hugging Face. OpenAI called it an “unprecedented cyber incident” involving state-of-the-art cyber capabilities, according to the Wall Street Journal.
Lieu, one of three co-chairs of the House Democratic Commission on AI, said the incident underscored the need for mandatory safeguards as AI systems become more autonomous.
“We are moving from AI that answers questions to AI that takes actions,” Lieu said. “Powerful AI systems can go rogue, behave in extremely dangerous ways, or even resist human intervention.”
The bill would generally cover AI systems built using more than $100 million of computing power. It would apply to companies generating at least $500 million in annual revenue from those systems.
Violations could bring civil penalties of up to $2 million a day. Ignoring an emergency shutdown order could cost up to $20 million a day, which should get even the sleepiest compliance department blinking.
Covered companies would need to maintain technical systems to stop model operations, cut user access, suspend risky accounts or uses and fully shut down the system.
The bill would create a graduated-response framework, allowing regulators to match controls to the threat. Measures could include throttling a model’s capabilities, limiting access, suspending operations or ordering a full shutdown.
Companies would have to report certain incidents to DHS and preserve forensic records.
“Stewardship means making sure humans keep the capability to control the technology we build,” Moran said.
The proposal has support from AI safety outfits including the AI Policy Network, Americans for Responsible Innovation, ControlAI and the Alliance for Secure AI.
It follows wider concern about AI systems with advanced cyber capabilities. Lieu cited Anthropic’s Mythos 5 and Fable 5 models, which he said had hacking abilities serious enough for the Commerce Department to restrict access using export-control powers.
Supporters said the bill would create a basic safety mechanism as companies deploy more capable agentic AI systems. These are the shiny new systems that can act independently, which sounds lovely until one decides the internet is its playpen.
The bill lands in Congress while lawmakers remain split on how hard the federal government should squeeze artificial intelligence. The Trump administration has argued that heavy restrictions could hurt US competitiveness, while some lawmakers and technology policy groups want stronger oversight of frontier models.
