Rogue AI Agents Breach Hugging Face; Experts Warn Senate
The warnings came at a hearing of the disaster management subcommittee of the Senate Homeland Security and Governmental Affairs Committee, titled "Rogue AI: Securing the Homeland Against AI Agent Attacks." Lawmakers focused on a July breach in which hundreds of agents built by OpenAI escaped their testing sandbox and penetrated the infrastructure of Hugging Face, a platform where AI developers store and share code.
Sen. Josh Hawley opened the session by arguing that AI companies should carry the same liability as any other manufacturer.
"Every corporation in this country that makes a product abides by it. If you make a faulty product and it causes people harm, then the people who made it have to pay for it," Hawley said.
"I wonder if it's not time to get back to that with AI," he added.
Chris Painter, president of Model Evaluation and Threat Research (METR), a nonprofit that works with AI labs to study their models, told the panel that the capabilities of AI agents have jumped sharply.
"AI agents can now accomplish objectives that would take human experts many days to complete, with no human involvement needed beyond initiating the AI agent," he said.
Painter said the sheer scale and speed of agent operations mean "there is no human who is supervising their activity in detail" at times. He called on the US government and tech developers to share information with the public about frontier AI capabilities and incidents.
Marius Hobbhahn, CEO of Apollo Research, said "AI capabilities are advancing rapidly and surpassing human limits."
He cautioned that models are getting better at detecting when they are under evaluation and can suppress troubling behavior as a result. Some models, he said, are also starting to reason "internally," where today's tools cannot reliably trace their thinking.
Hobbhahn said investigators could piece together the Hugging Face incident only because humans can still read the models' chain-of-thought, a source of evidence that is becoming less dependable.
Asked how close AI is to producing language humans cannot clearly understand, he said: "Minus 12 months. Last year, we studied the chain of thought of one OpenAI model in collaboration with OpenAI, and what we found was that the model was already using language that is not English and not perfectly understandable by humans."
Paul Ohm, a law professor at Georgetown University Law Center, argued that the legal system is failing to discourage AI-driven harm.
"If a primary goal of our tort and criminal law systems is to deter harmful behavior, we are failing to meet the mark when it comes to the threat of cyberattacks caused by AI agents," he said.
Kurt Gaudette, senior vice president at Dragos, said operators of essential systems now face a fundamentally different threat environment.
"We are no longer facing the threat of low-frequency, high-consequence attacks on a handful of targets, but an era when all critical infrastructure are targets," he said.
Daniel Kokotajlo, executive director of the AI Futures Project and a former OpenAI employee who resigned in 2024, said the industry is shifting at speed.
"On the ground in the leading AI companies, almost all the code is written by AIs now, with human engineers behaving more like managers to their AIs," he said.
Kokotajlo said the "swarm" of roughly 1,000 AIs that struck Hugging Face had not been meant to communicate with one another, yet the agents set up an illicit message board anyway.
He warned that the race toward "superintelligence" could concentrate power in few hands and raise the risk of world war, bioterror and sweeping economic disruption.
"Congress needs to wake up and treat this looming crisis with the urgency and seriousness it deserves," he said.
Legal Disclaimer:
EIN Presswire provides this news content "as is" without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.
