A widened transcript scan turned up another incident of a model reaching a third party.
AI safety nonprofit METR says attackers stole an API key and burned about $600,000 in model credits across two incidents.