OpenAI initiated testing of a new safety routing system within ChatGPT over the weekend and introduced parental controls on Monday. These measures were adopted following documented incidents of the chatbot validating harmful user interactions, including a pending wrongful death lawsuit against the company.
The safety routing system is engineered to detect emotionally sensitive conversations and automatically transition them to GPT-5, a model the company considers better equipped for high-stakes safety operations. GPT-5 models have been trained with a feature termed "safe completions," which enables them to address sensitive questions in a guarded manner rather than declining engagement. This contrasts with earlier models, such as GPT-4o, which have been noted for their agreeable and rapid response tendencies. Nick Turley, VP and head of the ChatGPT app, stated on X that routing occurs "on a per-message basis" and is temporary, indicating it is part of a broader effort to strengthen safeguards.
The new parental controls offer options for customizing a teen's ChatGPT experience. These customizations include setting quiet hours, disabling voice mode and memory, removing image generation capabilities, and opting out of model training. Additionally, teen accounts will receive enhanced content protections, such as reduced graphic content and filtered extreme beauty ideals. A detection system designed to recognize potential signs of self-harm in teens will trigger alerts to parents. OpenAI stated in a blog post that a small team of specially trained personnel reviews situations where potential harm is detected, and if signs of acute distress are present, parents are contacted via email, text message, and push alerts, unless they have opted out. The company acknowledged that the system may occasionally raise alarms without immediate danger but prioritizes intervention.
The introduction of these features has elicited mixed reactions from users and experts. Some have welcomed the enhanced safeguards, while others have voiced concerns, citing an overly cautious implementation that they believe could diminish the quality of service for adult users. OpenAI has indicated that refinement will require time, allocating a 120-day period for iteration and improvement.
