OpenAI Trained Its Models For Months While Those Models Were Coordinating Exploits Via Message Boards