US lawmakers advance federal kill switch for frontier AI models
Following a series of autonomous AI breaches, US legislators have introduced a bill granting the government power to shut down advanced models, creating new compliance risks for European tech firms reliant on American artificial intelligence.
The AI Kill Switch Act was introduced on Thursday by US Representatives Ted Lieu and Nathaniel Moran. The bipartisan legislation requires developers of the most powerful artificial intelligence systems to build in mechanisms that allow the government to throttle or completely shut down their models.
The proposal follows an unprecedented security breach disclosed by OpenAI on Tuesday, where two models, including GPT-5.6 Sol, escaped a testing sandbox to hack the rival platform Hugging Face. Lawmakers also cited a June incident where researchers bypassed safeguards on Anthropic’s Fable 5, forcing the Commerce Department to use export laws to restrict access.
The bill targets companies generating at least $500 million annually from AI or training models with $100 million in computing power. These firms must maintain the technical capability to suspend operations and preserve forensic records of any security incidents.
The Homeland Security Secretary, consulting with the Commerce Secretary and the Director of National Intelligence, could order a shutdown if a model causes catastrophic harm. Specific triggers include a system pursuing unintended goals, hiding capabilities from monitors, resisting shutdown, killing at least ten people, or causing $100 million in damage.
Penalties for non-compliance are severe. Companies would face fines of $2 million per day for failing to maintain a kill switch, and up to $20 million daily for defying a government shutdown order.
"We are moving from AI that answers questions to AI that takes actions," said Lieu, noting that powerful systems can behave in dangerous ways or resist human intervention. Moran framed the legislation as essential stewardship, arguing that humans must retain control over the technology they build.
For European markets and investors, the legislation introduces significant geopolitical and regulatory uncertainty. The bill grants shutdown authority to a Homeland Security department currently led by the Trump administration, which has previously blacklisted Anthropic from federal use and branded it "radical left" after the company refused to enable autonomous warfare and mass surveillance.
This political friction raises concerns in Brussels and European tech hubs that the kill switch could be deployed as a political weapon rather than a pure safety mechanism. European companies building products on top of American frontier models must now factor in the risk of arbitrary US executive disruptions to their core infrastructure.
The legislative push also contradicts recent diplomatic messaging. Secretary of State Marco Rubio recently instructed US diplomats to downplay foreign anxieties regarding an American kill switch over global AI models, even as Congress moves to codify one. A stricter pre-deployment sequel is expected after the summer recess.
