OpenAI has also faced scrutiny after an AI agent crossed testing limits by exploiting unknown weaknesses.