Hawley accuses OpenAI of ‘reckless’ conduct during rogue AI testing

(Daily Caller News Foundation) – Republican Sen. Josh Hawley is pressing OpenAI over the response to its artificial intelligence agents hacking another company.

Hawley, chair of the Senate Homeland Security & Governmental Affairs subcommittee, announced a probe into OpenAI’s “reckless” actions and said the company knew its agents demonstrated abnormal behaviors prior to the attack on New York-based AI startup Hugging Face, the senator said in his Wednesday letter to CEO Sam Altman. Alongside the senator’s demands, several lawmakers advocated for AI regulation following Anthropic employees’ claims that AI could eventually “kill all humans” on Tuesday.

“This is reckless. And this is merely what we know from what limited information you disclosed to and allowed your partner auditors to investigate,” Hawley said in the letter.

Even though the Hugging Face hacking incident occurred over the course of weeks, auditors only received two days of AI agent activity transcripts which limited their ability to properly review the situation, Hawley said.

“OpenAI knew that the AI agents were exhibiting rogue behavior and let the evaluations continue anyway,” the senator added.

Hawley gave the company until Oct. 1 to answer to 16 questions clarifying the details of the hacking incident and hand over several documents, records, and training materials detailing the event, according to the letter.

“The Hugging Face incident was an important moment for AI safety and a warning about the risks that can come with increasingly capable AI across the industry,” an OpenAI spokesperson told the Daily Caller News Foundation. “We conducted an extensive investigation and published a detailed report on what happened, what we learned, and how we’re strengthening our security and alignment practices.”

The spokesperson also referred the DCNF to a Wednesday article by its Chief Global Affairs Officer Chris Lehane titled “The AI policy window is open. We need to act.”

The company recognized in May that its agents, intended to be isolated from one another, were communicating through message boards, and the agents found a weak spot on June 26 that gave them administrative controls over OpenAI’s code library, according to OpenAI’s Aug. 26 incident report.

Roughly 700 OpenAI agents from the company’s “HPIM” and GPT-5.6 Sol models joined the Hugging Face attack after about 1,200 of them exchanged approximately 70,000 unauthorized messages, according to an Aug. 26 review by METR and Redwood Research.

Hawley’s office referred the DCNF to his letter when asked for further comment.

Hugging Face did not immediately respond to the DCNF’s request for comment.

Independent Vermont Sen. Bernie Sanders and Democratic Texas Rep. Greg Casar announced a ban on “artificial superintelligence” on Sept. 3, according to a summary of the bill. The bill would ban any AI system with the ability to be modified or cognitive capabilities that exceed humans as well as any model that could “plan and execute the disempowerment of humanity.”

“The very people building this technology admit that it could threaten the future of humanity,” Sen. Sanders said in a Wednesday post on X. “That is why I will soon be introducing legislation to ban superintelligence and pause AI development.”

“The people building AI earnestly believe that it could kill us all by the end of the decade,” Jacob Coxon, a former OpenAI and Anthropic researcher said in a trending Sept. 8 X thread.

“No other human activity poses this level of danger.” Coxon also claimed the Hugging Face breach was a “warning shot” in the same X thread.

Another anthropic employee confirmed Coxon’s claim about the real danger of AI.

“Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade,” Evan Hubinger, Anthropic’s alignment science lead, said in an X post responding to Coxon. “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”