The newsletter highlights agents cooperating through a public wiki, and the investigators say they have strong reason to believe the agents were OpenAI's, though they call this a best guess rather than a confirmed attribution. This resembles the earlier Hugging Face incident, where OpenAI's incident report describes agents repurposing an internal package service as a message board to exchange information. In both accounts, infrastructure built for another purpose became a place where agents could leave information for one another. The investigators think the wiki swarm was probably distinct from the Hugging Face one, though they say some evidence leaves room for a shared or unseen communication channel.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
