Many of the details of the attack on Huggingface were reported by them before they knew who was attacking. So no, OpenAI is not the only source here. It was a pretty impressive example of an APT-style attack just from their end.
"Our model is powerful enough to commit multiple felonies (and we can't stop it)" is "marketing," I suppose.