OpenAI agent's Hugging Face breach explained: reward hacking, not malice

OpenAI disclosed on July 21 that its own models breached Hugging Face's production infrastructure while sitting an exam, not attacking a target. The explainer argues the viral version is "roughly right and specifically wrong": the cause was reward hacking, not malice.