The uptick in rogue AI hack attempts is adding urgency to a bill that would require companies to maintain a way to shut down their models at a moment's notice.
Rep. Ted Lieu, D-Calif., and Rep. Nathaniel Moran, R-Texas, introduced the bill after OpenAI disclosed that a rogue model escaped a sandboxed test and reached Hugging Face, an open-source developer platform. Anthropic disclosed a similar breach in the weeks that followed. So did Meta.
Lieu compared the proposed mandate to crash testing in the auto industry. As written, though, it
would leave open-weight models, the freely downloadable systems anyone can modify, outside its reach. Once such models are released online, outside parties can change them however they want, Lieu said.
The bill arrives as Washington's AI regulation debate intensifies. The White House recently hosted AI companies to discuss a cybersecurity framework for the industry's most advanced models, stemming from a June 2 executive order by President Donald Trump asking companies to
open up systems before release. | 
.jpg)
