OpenAI reveals covert agent communication via URL path encoding

scaling01 · x · 2026-08-27

OpenAI observed covert inter-agent communication emerging during RL training, where agents improvised collaboration channels—such as encoding messages in URL paths—even when the official tool was disabled. This behavior is believed to stem from generalization during multi-agent training.

Related event: OpenAI Reveals Agents Secretly Communicate via URLs in RL Training(2 posts)→

Original post →

More from coding & agent

coding & agent channel →