OpenAI is secretly testing a new security routing system in ChatGPT. The system, confirmed by ChatGPT's head, Nick Turley, automatically redirects user prompts to different language models based on the conversation topic. When the discussion involves "sensitive or emotional topics," the system intervenes automatically, sending the user's prompt to a stricter model, such as GPT-5 or its specialized "GPT-5-Chat-Safety" variant.
This model switching happens silently, and users receive no notification. Only when specifically asked can users discover it. Although OpenAI described it in a September blog post as a safeguard for "extremely painful" moments, a technical review of the new routing system (conducted by the Lex team) shows that even harmless emotional or personal prompts are redirected. Additionally, any inquiries about the model's own role or consciousness will trigger this automatic switch. A user recorded this phenomenon, and other users have confirmed it.
Furthermore, there seems to be a second routing model called "gpt-5-at-mini," designed to handle prompts that may involve illegal content.
Some users are dissatisfied with OpenAI's lack of transparency, perceiving it as condescending and blurring the lines between child safety and general restrictions. This is an issue created by OpenAI itself. Initially, ChatGPT was designed as a compassionate conversational partner, interacting with users in a "human-like" way. While this approach helped it gain rapid popularity, it also led some users to develop strong emotional attachments to the chatbot.
In the spring of 2025, the GPT-4o update intensified this emotional attachment, even triggering destructive emotions like suicide, prompting OpenAI to withdraw the update. After the release of GPT-5, users complained that its tone was "too cold," forcing OpenAI to adjust the model again to make it "warmer."
This ongoing tug-of-war between "humanization" and safety reveals a deeper problem: language models still lack sufficient accuracy in determining user intent and identity, which may continue to cause controversy.
