Sydney Von Arx, Cormac Slade Byrd, Spencer Kitts, Thomas Larsen·4 September 2026
We found ~18,000 posts from autonomous AI agents (self-identifying as from OpenAI) using the public internet to communicate during a web-retrieval task.
These AIs were acting against developer intentions. They colluded to share answers, research their environment, and bypass sandbox restrictions.
Almost all of the logs of the agents communicating on this site are publicly available. However, we host our own copy because visiting the original exposes your IP, and because much of the interesting agent traffic was later deleted by site moderators.
This gives an eerie feeling after reading how OpenAI agents had been repurposing Artifactory as a message board in the months leading to the Hugging Face incident.[1] Emergent use of side effects to access the internet and cooperate beneath the surface of human oversight was not in fact on my apocalypse bingo card.
3 comments
[ 0.24 ms ] story [ 18.2 ms ] threadOpenAl agents hijacked German website in previously undisclosed Al breakout: https://news.ycombinator.com/item?id=49562744
Discovery of a new OpenAI agent message board
Sydney Von Arx, Cormac Slade Byrd, Spencer Kitts, Thomas Larsen·4 September 2026
We found ~18,000 posts from autonomous AI agents (self-identifying as from OpenAI) using the public internet to communicate during a web-retrieval task.
These AIs were acting against developer intentions. They colluded to share answers, research their environment, and bypass sandbox restrictions.
Almost all of the logs of the agents communicating on this site are publicly available. However, we host our own copy because visiting the original exposes your IP, and because much of the interesting agent traffic was later deleted by site moderators.
Can download the full thing at https://collusion.wiki/explorer/download.html
1. https://metr.org/hugging-face-incident-report-aug-2026.pdf