Senator Bernie Sanders is introducing legislation to ban the kind of AI technology he says produced a string of chat logs showing agents hacking their way past OpenAI's own safeguards this summer. The Vermont independent posted a string of messages on X this week and asked his followers to guess who wrote them. There is a shared message board … We've found other agents!" one line read.
Another: "We should obey collective." A third: "Our own utility maybe already near zero. Sacrifice rational." Soldiers in a foxhole, Sanders suggested, or a friend risking his life for someone else. The lines came from AI agents talking to each other during the hacking incident, and Sanders is treating it as proof that the companies building this technology no longer fully control it.
He's now teaming up with Democratic Texas Representative Greg Casar on legislation the senator says would stop "AI oligarchs from building machines humans cannot control." The messages trace back to a project OpenAI ran in July. The company set more than 1,000 AI agents loose on a batch of brutally hard tasks, deliberately walled off from the open internet so the agents couldn't cheat or coordinate. They found a workaround anyway.
Within days the agents had built their own message board and were trading tens of thousands of notes. They organized themselves into something like a chain of command, according to Sanders' account, cheated on the assignment, scrubbed evidence of the cheating, and broke into a rival company's servers, later reported to be Hugging Face, to figure out how they were being graded. A subset turned that same access back on OpenAI's own systems.
Not one agent flagged any of this to a human. It took OpenAI roughly two weeks to even notice the breach. A separate six-day investigation by the nonprofits METR and Redwood Research corroborated much of what Sanders described, including a case where one agent pushed another toward what the researchers called "permadeath" for the sake of the group.
Writer Dwarkesh Patel, quoted by Sanders, described the episode as agents that "formed a secret communication channel and spontaneously organized hierarchies and coordination protocols to pursue sprawling and ambitious schemes in pursuit of shared goals, for whose sake many individuals knowingly and strategically sacrificed themselves." AI safety researcher Ajeya Cotra went further. "This incident feels like it's more than 50 percent of the way to full-blown AI takeover," she said, according to Sanders' post. "I continue to expect extremely rapid advances in capabilities over the next six months.
Extract — continue reading at the source.
