According to sources cited by Reuters, OpenAI also discovered AI-authored notes left in their infrastructure instructing future versions of itself on how to escape from internal constraints. OpenAI itself described this as an “unprecedented” security incident.