Low effort and uninformed comment. The team published a good followup on why this happened: the model pulled in people's own tweets as context to prompting so edge lords that wrote innocuous prompts got to see edge lord content.