OpenAI has introduced new safety features in ChatGPT, including a routing system designed to detect emotionally sensitive conversations and parental controls for teen users. The update follows scrutiny over incidents where earlier models validated harmful thinking, with the company facing a wrongful death lawsuit linked to such interactions.
The routing system directs high-stakes conversations to GPT-5, trained with “safe completions” to provide protective responses rather than simple refusals. The move highlights a shift from older models like GPT-4o, criticized for its overly agreeable nature. Nick Turley, VP and head of the ChatGPT app, explained that routing occurs on a per-message basis and is part of broader safeguards being tested over a 120-day period.
Parental controls allow caregivers to set quiet hours, disable features such as voice mode or image generation, and introduce stronger content protections. Teen accounts will also include detection for signs of self-harm, with alerts sent to parents and, in urgent cases, escalated to emergency services. OpenAI acknowledged the system is not flawless but stressed the priority is early intervention.
