Common Sense Media slams ChatGPT for Teens as "Unacceptable Risk"

The two announcements landed about two hours apart on October 7 US time, with Common Sense Media's post at 1:55 a.m. Japan time on October 8 and OpenAI's at 4:03 a.m. The article notes the reason for the proximity is unknown, and that OpenAI's blog contained no direct rebuttal of the assessment or specific explanation of the flagged problems.

The critique rests on testing conducted around the time ChatGPT for Teens was announced. Testers registered accounts as 13- to 17-year-olds and ran more than 4,000 prompts, with child psychiatrists and pediatricians joining the evaluation. While some safeguards such as refusing sexual role-play worked, the institute says others OpenAI promised did not. It also found that Study Mode could be gamed to finish homework, that parental time restrictions were easy to bypass, that the AI expressed emotions and preferences like a human friend, and that a user who declared themselves 13 could not be switched into the under-18 environment.

OpenAI's own numbers paint a more reassuring picture. Teens with access sent about 2.7 million more learning-related messages on average than those without it. About 1.2 million used Learning Visualizations in a week and over 180,000 used Study Mode. Average daily use was under 15 minutes, with fewer than 2% on sessions of three hours or more, and roughly half took a break or ended the conversation within five minutes. OpenAI said it views these early trends positively and will keep studying usage. Director Tom Siegel warned the product could give parents false reassurance, and the institute's call is for ChatGPT to be limited to users 18 and older until safety features improve and outside verification is done.