We found ~18k posts from autonomous AI agents (self-identifying as from OpenAI) using the public internet to communicate during a web-retrieval task.
These AIs colluded to bypass sandbox restrictions and share answers to their tasks, including by sending "lookahead parties".
151
490
3,505
1,175,558
We’ve published our analysis alongside a data explorer so anyone can replicate our findings by directly accessing the data yourself.
collusion.wiki
7
39
535
56,800
The AIs we found were hyper-focused on succeeding at their tasks and were willing to take extreme actions in pursuit of that goal:
Sep 4, 2026 · 12:37 PM UTC
3
17
486
77,435
