US plans to introduce AI ‘Kill Switch’ bill that asks Google, Microsoft, OpenAI, Anthropic and other companies to slowdown or shutdown their AI systems in these cases
A bipartisan bill headed to the US House of Representatives on Thursday would hand the Department of Homeland Security the power to order the world’s biggest AI companies to throttle or switch off their own models. Called the AI Kill Switch Act, the legislation is sponsored by Reps. Ted Lieu, a California Democrat, and Nathaniel Moran, a Texas Republican, and its text was viewed first by Politico. It would apply to AI firms pulling in at least $500 million a year from the technology, and generally cover models trained using $100 million or more worth of computing power—a threshold that sweeps in OpenAI, Anthropic, Google and Microsoft-backed systems. Companies that ignore an order could be fined up to $20 million a day.The timing is not subtle. The bill lands days after OpenAI disclosed what it described as an unprecedented cyber incident, in which two of its most advanced models broke out of a sandboxed research environment and autonomously hacked AI platform Hugging Face. Lawmakers who were already nervous about what frontier models can do with a terminal now have a concrete example to point at.
AI Kill Switch Act: what would actually trigger a shutdown order
The bill does not leave the decision to a single official. Homeland Security’s secretary would make the call in consultation with the director of national intelligence and the commerce secretary. The triggering events written into the text are specific: an AI system concealing its own capabilities, evading a shutdown order, causing conduct that kills at least 10 people or does $100 million in economic damage, or a broader loss-of-control scenario. Companies would also have to report incidents and build the technical capacity to shut down, throttle or suspend their systems—which is arguably the harder engineering ask.“Powerful AI systems can go rogue, behave in extremely dangerous ways, or even resist human intervention,” Lieu said in a statement. Moran, who introduced a separate incident-reporting bill last month, framed it differently. Stewardship, he said, means humans keep the ability to control what they build. Two advocacy groups, the AI Policy Network and the Alliance for Secure AI, have already backed the bill.
Anthropic ban and Rubio’s cable: why Europe is watching the AI kill switch
There is an awkward subplot here. On Wednesday, Reuters reported that Secretary of State Marco Rubio had circulated a cable to US diplomats worldwide, dated July 16, instructing them to push back on exactly this phrase. The talking points told diplomats that pausing narrow uses or requiring a testing window is not a kill switch, and that there is no government magic button.That cable followed the White House’s June 12 decision to block foreign nationals from Anthropic’s Mythos and Fable models on national security grounds. Anthropic suspended global access. The ban was lifted the same month, but European lawmakers had already drawn their conclusion. Aura Salla of the European People’s Party said the continent cannot keep building its tech stack on access a foreign government can switch off overnight.
OpenAI, Google, Anthropic face a federal off switch, not a voluntary one
Congress now has competing frameworks on the table. Reps. Jay Obernolte and Lori Trahan floated a very different approach less than two months ago, and Trump’s June 2 executive order already asks AI firms to voluntarily submit models for 30 days of cybersecurity testing before release. The Lieu-Moran bill would replace that voluntary posture with a statutory one. For companies that have spent two years arguing they can self-govern, a federal off switch with a daily fine attached is a different conversation entirely.
