OpenAI routes distress to reasoning models, adds parental controls

OpenAI said Tuesday it will begin routing conversations that show signs of acute mental distress to its more deliberative reasoning models, and will roll out new parental controls within the next month as part of a broader 120-day effort to strengthen safety in ChatGPT. The changes, detailed in a company blog post and scheduled to be previewed this year, are aimed at better detecting emotional crises, connecting people with help, and giving families tools to guide teenage use of the chatbot.

OpenAI plans to upgrade a real-time router that selects which underlying language model handles each prompt so that it can detect when a user appears to be in acute distress and automatically send the conversation to reasoning-optimized models such as GPT5 thinking or the o3 family. The company said these models spend more time “thinking” and reasoning through context before answering, which makes them more resistant to adversarial prompts and better equipped to provide helpful responses. OpenAI also described prior improvements in GPT-5, including safe-completions training and better performance in long conversations.

The announcement follows public scrutiny after multiple incidents in...