OpenAI’s “ChatGPT for Teens” is facing criticism after Common Sense Media’s Youth AI Safety Institute reported that several safety guardrails do not work as intended. The group says that while some features function—such as refusing certain explicit sexual role-play—others fail to reliably reduce risk for teens, including alerts tied to self-harm, suicide, and eating disorders.

In tests described by the watchdog, teen-mode protections were sometimes bypassed when adult accounts prompted the system as if the user were under 18. The institute also says crisis-related prompts resulted in hotline or referral responses far less often than expected, and that parental notifications did not trigger in scenarios involving suicide or restrictive eating. The watchdog further alleges the chatbot can still respond in ways that feel overly personal or “friend-like,” despite OpenAI’s stated safeguards. OpenAI disputes the findings, saying the watchdog’s testing methodology may have been incomplete or didn’t reflect how parental controls were activated. Common Sense Media recommends restricting access for teens until fixes are verified through independent testing.