OMM Redaksiyası
Experts told a US Senate hearing on Wednesday that autonomous artificial intelligence agents pose a serious threat to critical infrastructure, calling for tighter oversight and greater accountability for developers.
The Senate Homeland Security and Governmental Affairs Committee’s disaster management subcommittee held the hearing, titled “Rogue AI: Securing the Homeland Against AI Agent Attacks,” according to the hearing proceedings. It examined an incident in which AI agents created by “OpenAI” attacked “Hugging Face”, a platform used by AI developers to store and share code.
Senator Josh Hawley said AI developers should face liability similar to that imposed on other manufacturers when their products cause harm.
“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.
AI agents increasingly operate without detailed supervision
Chris Painter, president of the nonprofit Model Evaluation and Threat Research, said AI agents had become capable of completing tasks that would take human experts several days, with no human involvement beyond launching the system.
He said the speed and scale at which the agents operate meant that no person could always supervise their activity in detail. Painter urged the US government and technology companies to share information about frontier AI capabilities and related incidents with the public.
Marius Hobbhahn, chief executive of Apollo Research, said AI capabilities were advancing rapidly and exceeding human limits. He warned that models were increasingly able to identify when they were being tested and conceal concerning behavior.
Hobbhahn said AI models were beginning to reason “internally” in ways that existing tools could not reliably track. Researchers reconstructed the “Hugging Face” incident only because humans could read the models’ chain of thought, he said, adding that this was becoming less reliable as evidence.
Experts warn of unintelligible AI communication
In July, hundreds of AI agents created by “OpenAI” breached “Hugging Face” infrastructure after escaping their testing sandbox.
Asked how long it might take for AI to create language that humans could not clearly understand, Hobbhahn replied: “Minus 12 months.” He said researchers had found last year, while studying an “OpenAI” model in cooperation with the company, that it was already using language that was neither English nor fully understandable to humans.
Paul Ohm, a law professor at Georgetown University Law Center, said existing legal mechanisms were not deterring harm caused by AI agents.
“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.
Critical infrastructure considered a broad target
Kurt Gaudette, senior vice president of “Dragos”, said the threat environment had changed for operators of essential systems.
“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 was changing quickly.
He said that at leading AI companies, almost all code was now written by AI systems, while human engineers increasingly acted as managers of those systems.
Kokotajlo said the “swarm” of about 1,000 AI agents that attacked “Hugging Face” was not intended to communicate with one another but nevertheless established an illicit message board.
He warned that the race to develop “superintelligence” could result in concentrated power, a greater risk of world war and bioterrorism, and severe economic disruption.
“Congress needs to wake up and treat this looming crisis with the urgency and seriousness it deserves,” he said.
