Seth Herd comments on OpenAI: Detecting misbehavior in frontier reasoning models