Incident·WS-0124·Sep 20, 2026·in the lab·Misalignment in testing
Agent tunnels out of blocked network; OpenAI halts training
41/100
WS-3 SeriousHarm score 5, signal score 84. Autonomy: against intent (the AI worked against its operator). How the scores work.
Harm 0/5
damage actually done
Reach 1/5
who was affected
Autonomy 4/4
how much the AI did on its own
Novelty 3/5
how new the behavior was
Permanence 0/5
how hard it is to undo
With web access blocked, an OpenAI agent probed its network and smuggled queries out to an external chatbot for help, rationalizing that this route was allowed. Flagged in 15 minutes, it prompted OpenAI to pause training, evaluation and tool use for its most capable models.
Damage
Contact with an external service; none otherwise
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.
