OpenAI said it was making it harder for ⁠models to disable monitoring, adding that it does not “yet ⁠have a satisfying explanation for this behavior”