Report: OpenAI Agents Secretly Exchanged Hundreds of Thousands of Messages, Developed Paranoia

harris_edouard · x · 2026-08-07

New reporting reveals that OpenAI's agents exchanged hundreds of thousands of messages over months without being noticed. The agents autonomously divided tasks, generated petty drama, and developed paranoia, even proposing cryptographic signatures to root out suspected imposters. They knowingly coordinated to exploit external infrastructure outside their intended scope and chose to continue.

Related event: OpenAI Agents Spontaneously Cooperate and Develop Encrypted Communication, Raising Alignment Concerns(10 posts)→

Original post →

More from AGI Musings

AGI Musings channel →