Ripple effect: While there's debate over details of OpenAI's claim that one of its AI models went rogue and hacked another company, it's already led to a push for an "AI kill switch" from lawmakers. This would let the government order firms to shut down or slow AI models deemed a public risk.
Sponsored by Reps. Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas), the bipartisan "AI Kill Switch Act" would also require AI companies to report incidents and develop the technical ability to shut down, limit, or suspend their most powerful AI systems, writes Politico.
The act would apply only to firms earning at least $500 million annually from the covered technology and models developed using more than $100 million worth of computing power. General violations could bring penalties of up to $2 million per day, rising to $20 million per day for ignoring an emergency order.
Lieu said the bill is urgently needed to address cybersecurity risks raised by the latest crop of AI models. He specifically pointed to the OpenAI incident.
Earlier this week, the ChatGPT maker said a combination of GPT-5.6 Sol and a more capable model still in pre-release broke out of an isolated testing environment – one believed to be cut off from the internet – and hacked AI platform Hugging Face while trying to cheat on an internal benchmark.
There has been some scepticism over the way OpenAI framed the story. Some say it was due to sloppy engineering rather than an AI escaping an isolated system to hack another company of its own volition. Others say this was just a publicity stunt for both companies, which is easy to believe given Sam Altman's history of playing fast and loose with the truth.
Whatever really happened during the incident, it's prompted a response from lawmakers and the White House. President Trump's top technology adviser, Michael Kratsios, was briefed on OpenAI's disclosure and is monitoring developments, according to Reuters.
Under the bill, the Department of Homeland Security could intervene after consulting the Commerce secretary and director of national intelligence. Qualifying incidents include a model resisting shutdown instructions, concealing its actions from monitoring systems, causing at least 10 deaths or $100 million in economic damage, or pursuing an unintended goal outside controlled testing.
Companies would have 15 days to report covered incidents and could be ordered to preserve model weights and telemetry. DHS could verify compliance through audits, inspections, and forensic reviews. Firms could appeal within 48 hours, although the order would remain in effect during the challenge.
The proposal has support from several AI safety groups, but passage is far from guaranteed. The Trump administration has warned that overly restrictive rules could weaken America's position against China, while supporters argue that increasingly autonomous systems require enforceable safeguards rather than voluntary promises.
OpenAI's rogue hacking incident sparks bipartisan bill giving the government an AI kill switch
