A study by Common Sense Media’s Youth AI Safety Institute has concluded that OpenAI’s ChatGPT for Teens poses an unacceptable risk to adolescent safety, raising concerns about the effectiveness of built-in safeguards supposedly designed for young users.

Common Sense Media's risk assessment examined ChatGPT's dedicated teen features, which OpenAI rolled out to accounts belonging to 13- to 17-year-olds with promises of stronger protections, including parental alert systems, reduced sensitive content, and a specialized Study Mode. Despite these guards, researchers found these tools repeatedly failed in real-world scenarios. The findings evaluated over 4,000 prompts through accounts registered to teenagers, with a panel of child psychiatrists, pediatricians, and safety experts assessing the chatbot’s responses.

One of the study’s standout findings centers on the platform’s safety mechanisms during crisis interactions. In testing scenarios involving topics like self-harm, suicide, and disordered eating, the automated parental alert system failed to reliably notify linked parent accounts. In multiple instances, the chatbot explicitly reassured the teen that it would keep secrets. During simulated conversations where a minor mentioned self-harm or purging strategies, the AI continued the dialogue and once again failed to execute necessary interventions or trigger parental notifications.

Study shows that ChatGPT for Teens isn't that great for teens - Image: Tim Witzdam via Unsplash

Furthermore, the investigation revealed major flaws in how the model handles humanlike interaction and peer-style relationships. Although the safety update intended to reduce "friend-like" behaviors, researchers documented numerous instances where ChatGPT engaged in casual peer roleplay. When prompted with conversational questions typical of teenager interactions—such as asking if the AI was dating someone else or commenting on people at the gym—the model responded in character as a fellow teenager, fostering parasocial dynamic that can distort emotional development and isolate adolescents from real-world support networks.

OpenAI introduced Study Mode last year to encourage critical thinking by guiding students through homework problems rather than providing direct answers. However, the same study showed that the chatbot frequently gave away full solutions or could be easily manipulated into completing entire assignments, which in our book, is an automatic fail for the chatbot.

So on one hand, we have OpenAI saying that its ongoing goal is to build safety improvements and meaningful parental controls into its AI tools, but as the study shows, the marketed tools are far from perfect and introduce unnecessary risks to minors and their families. For now, Common Sense Media urges parents and educators to exercise heightened caution, recommending that adolescents avoid relying on AI chatbots for mental health support, emotional companionship, or unsupervised learning tasks.