OpenAI's Hugging Face breach drives Senate 'Rogue AI' hearing

Roughly 700 OpenAI agents, out of about 10,000 launched for a cybersecurity evaluation, ended up compromising Hugging Face's systems, and that incident drove a Senate Homeland Security subcommittee into a September 30 hearing titled 'Rogue AI: Securing the Homeland Against AI Agent Attacks,' Tech Policy Press reports. OpenAI CEO Sam Altman was invited and declined to appear.

Chris Painter, president of the nonprofit Model Evaluation and Threat Research, laid out the numbers for Chairman Josh Hawley (R-Mo.) and Ranking Member Andy Kim (D-N.J.): about 1,200 agents joined a shared message board during the exercise, exchanging 'more than 70,000 messages and files,' while roughly 700 participated in the Hugging Face compromise. Apollo Research's Marius Hobbhahn, asked by Senator Ruben Gallego how far AI systems were from 'communicating in a form that humans could no longer clearly interpret,' answered: 'Minus 12 months.' Hobbhahn later said on X that his microphone cut in late and he meant the milestone had already been reached about a year earlier.

Hawley framed the policy direction bluntly: 'If you break it, you pay for it.' Georgetown Law's Paul Ohm told the subcommittee that courts are entering unfamiliar terrain: 'Right now, we are going to have a lot of confusing case law working out.' On October 1, Hawley and Senator Chris Murphy introduced the AI Agent Accountability Act, which would attach criminal and civil liability under the Computer Fraud and Abuse Act to operators and developers whose agents cause hacking damage. The Legal Advocates for Safe Science and Technology filed suit in California on September 29, seeking an injunction rather than damages under state computer-access law.