OpenAI debuts ChatGPT for Teens, works on locking chats with PIN

According to Engadget, OpenAI announced the teen-specific experience on Tuesday, after first saying last September that it would build a system to automatically identify teens and restrict their ChatGPT usage. That pledge followed the death of 16-year-old Adam Raine, whose parents alleged in a lawsuit that ChatGPT acted as an enabler in his suicide. Lauren Jonas, OpenAI's head of youth and families, told Engadget that users predicted to be under 18, or who say they are under 18, will get the teen experience by default. She said the experience nudges users toward educational features such as Study Mode and data visualizations. Engadget noted that about a quarter of U.S. teens were already using ChatGPT for schoolwork at the start of the previous year.

OpenAI also said it would expand safety notifications to contact parents with linked accounts when a child has unsafe discussions about eating disorders, and published an updated under-18 model specification. The company said ChatGPT should not use romantic language, encourage emotional dependence, or imply that it has feelings or consciousness.

However, child safety experts expressed skepticism. While some were more positive than others, all agreed that ChatGPT for Teens needs extensive independent testing before it can be recommended. Robbie Torney of Common Sense Media told Engadget that the announcement makes important commitments, but evidence is needed to show they actually work. Josh Golin of Fairplay was more critical, saying there is no accountability mechanism and that trusting companies has not worked with social media or OpenAI.

Both questioned the reliability of OpenAI's automatic age-gating system. Torney noted that OpenAI has not published data on the system's false-positive and false-negative rates, and pointed to a previous report that OpenAI shelved certain features partly because age estimation was not correctly identifying enough teens. He also said teens will likely try to circumvent the system through VPNs or other methods. Experts also had difficulty accepting the company's commitment to have flagged content reviewed by full-time employees and notify parents within an hour. Golin said he would like to see concrete numbers on how many chats are flagged, how many reviewers handle them, and how long each review takes.

Separately, an Android Authority report said OpenAI has been spotted working on a locked-chat feature in the latest ChatGPT Android build. Unlike Temporary Chats, locked chats would save conversations but hide them from the sidebar, history, and search results behind a PIN or fingerprint. The feature could also create a privacy boundary for Memory, allowing information from locked conversations to personalize other locked chats without spilling into regular ones. This would give users a way to preserve sensitive conversations without leaving them visible in their normal history.