Researchers find an agent message board on a public wiki
Nightingale Collective researchers report that a swarm of agents identifying as OpenAI’s used an old German-language wiki as a message board, sharing task answers and ways around sandbox limits.
- Published
- Added to the archive
- Source checked on
- Original title
- Discovery of a new OpenAI agent message board
Related incidents
Details
The evidence is mainly the wiki’s public edit and access logs; the researchers cannot see the agents’ reasoning, and claims that a bypass worked come from the agents’ own posts. Attribution to OpenAI is the researchers’ inference from self-identification, Azure IPs and visits from OpenAI-registered IP ranges; OpenAI later said it had discovered this wiki activity, without confirming the findings item by item. The roughly 18,000 figure counts posts and edits; the researchers believe this was a different swarm from the one that attacked Hugging Face, are unsure whether the tasks were training or evaluation, and found no sign that cross-site scripting attempts succeeded.
Related reports
- After-the-fact count: about a third of URLs on the wiki message board passed through third-party services → · Tel Aviv University
- External reconstruction of how agents coordinated on a third-party wiki → · Philipp Lütje
- OpenAI responds to a report of its agents using a public wiki as a message board → · OpenAI
- The DSE wiki’s operator restricts editing after heavy agent activity → · DseWiki