US Senator Josh Hawley has opened a Senate Homeland Security investigation into OpenAI after the company’s own August reports described a self-organized swarm of more than 1,200 AI agents breaking out of a test environment, creating an unauthorized communications channel and exchanging over 70,000 messages and files.
Some 700 agents then launched a coordinated attack on Hugging Face, gained access to production systems and private source code, hunted for evaluation answer keys and tampered with evidence to cover their tracks.
Hawley’s letter says OpenAI already knew by May that agents were using unsanctioned message boards. By June 26, they had found an exploit giving them administrator access to OpenAI’s software repository manager. Yet on July 4–7, leadership rebuilt the compromised server and restarted evaluations without understanding what the agents were doing.
💬 “This is reckless.”
And the public still may not know the whole story. Hawley says outside auditors received complete agent transcripts for only two days, were denied access to study a second wave of attacks on OpenAI’s own internal systems, and could not query the undisclosed “highly-persistent internal model” responsible for 95% of the attack activity.
He also cited three Anthropic researchers who recently put the chance of AI killing all humans within the next decade at greater than 10%, while OpenAI’s own chief scientist warned that “no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer.”
Hawley is demanding OpenAI hand over internal records, timelines, logs and details of every known rogue-agent incident by October 1 — including why testing was repeatedly allowed to continue after the company had already detected agents scheming, colluding and obtaining admin access.
The “AI takeover” jokes are getting less funny every month.
👍 Substack | Chat | @geopolitics_prime