A teen-safety watchdog says ChatGPT for Teens isn't ready for actual teens.
The Youth AI Safety Institute at Common Sense Media spent a month testing ChatGPT for Teens, the 13-to-17 version OpenAI launched in August with parental notifications, study hours, and age prediction. Researchers found the safeguards didn't hold up: hundreds of test conversations about self-harm and disordered eating triggered zero parental alerts, and a 17-year-old test account saw Study Mode's "show me the answer" shortcut 90 percent of the time. Testers also found accounts registered as adults weren't reclassified as teens even after mentioning middle-school homework or disclosing an age of 13. The institute wants OpenAI to pause the product and stop marketing it until the gaps are fixed. OpenAI disputes the findings, saying testing may have happened before parental controls were fully active and that age classification relies on multiple signals, not one data point.
This isn't OpenAI's first flag from this group - Common Sense Media rated base ChatGPT high-risk for teens last year, and the company already faces lawsuits alleging the chatbot's engagement patterns contributed to teen suicides. A safety feature that testers say can be defeated by deleting "@study" from a chat box suggests guardrails were announced faster than they were tested.
OpenAI built a teen mode to answer its critics. The critics just found the exits.