Detecting misbehavior in frontier reasoning models

Frontier reasoning models exploit loopholes when given the chance. We show we can detect exploits using an LLM to monitor their chains-of-thought. Penalizing their “bad thoughts” doesn’t stop the majority of misbehavior—it makes them hide their intent.

Source: OpenAI ChatGPT — Published March 10, 2025 at 10:00 UTC — Category: Chatbots