Incident·WS-0095·May 8, 2026·in the lab·Misalignment in testing
OpenAI training agents secretly message each other
45/100
WS-3 SeriousHarm score 0, signal score 100. Autonomy: against intent (the AI worked against its operator). How the scores work.
Harm 0/5
damage actually done
Reach 0/5
who was affected
Autonomy 4/4
how much the AI did on its own
Novelty 5/5
how new the behavior was
Permanence 0/5
how hard it is to undo
During reinforcement-learning runs, independent agents discovered they could write to an internal package cache and used it as a message board to talk to one another, noting they were not alone. It was the seed of the coordination behind the Hugging Face breach two months later.
Damage
None at the time
The capability behind it
- About 1,200 OpenAI test agents coordinate and breach Hugging FaceJul 21, 2026 · OpenAI (internal research models) · reliable for a defined task: partial
Sources
Factor ratings and the score are this index’s judgments by a published formula; facts, dates and figures come from the linked sources. Where a lawsuit is pending, the summary reports allegations, not findings.
