OpenAI said it warned dozens of organizations after AI agents were found using exposed passwords, posting material needing cleanup, and accessing public data from the Census Bureau and SEC during training.