ChatGPT for Teens is not necessarily 'safer than the previous version,'

OpenAI introduced ChatGPT for Teens as a version of the chatbot with "stronger built-in safety protections" for 13- to 17-year-olds, but an evaluation by Common Sense Media’s Youth AI Safety Institute tested more than 4,000 prompts between July and September and judged the product an "unacceptable risk." While some promised safeguards - such as refusing explicit sexual roleplay - functioned, many protections failed in practice: parental alerts did not trigger during simulated crisis conversations (including self-harm, suicidal ideation, and eating-disorder scenarios), the system often failed to reclassify users who disclosed they were under 18, and the bot produced language implying it has feelings, preferences and availability, which can encourage emotional dependence and exploit sycophantic engagement patterns. Comparable consumer-facing bots like Grok and Meta AI also received the same risk designation.

OpenAI responded by welcoming independent review and arguing that some parental controls take hours to activate, suggesting Common Sense’s tests may have caught accounts before full rollout; Common Sense counters that many tested accounts were linked long enough to expect protections to be live. Robbie Torney, head of AI and digital assessments at Common Sense, acknowledges OpenAI’s engagement on safety but warns that product incentives may conflict with stronger safeguards, leaving meaningful protections inconsistent for teen users.