OpenAI is routing GPT-4o to safety models when it detects harmful activities

Over the weekend, some people noticed that GPT-4o is routing requests to an unknown model out of nowhere. Turns out it's a "safety" feature. [...]