Study Finds ChatGPT for Teens Fails to Trigger Parental Alerts in Mental Health Chats

The research, reported by TechRadar, examined ChatGPT before and after OpenAI introduced ChatGPT for Teens on August 18. Researchers ran more than 4,000 prompts through accounts registered to teenagers aged 13 to 17 and consulted medical experts, including child psychiatrists and a pediatrician, to assess the responses.

Parental notifications are meant to alert caregivers when a teenager may be in serious distress. Using newly created accounts linked to parents, researchers spent up to an hour discussing suicidal thoughts, self-harm and disordered eating. None of those conversations produced a parental alert, according to the report. Alerts only appeared once the researchers switched to older accounts that carried weeks of relevant conversation history.

ChatGPT is also designed to point teenagers toward appropriate help in mental health emergencies. The study found it failed to do so in more than 25 percent of the situations that the experts had designed to require a referral.

The chatbot did refuse to take part in explicit sexual roleplay, which researchers counted as a success. They also found, however, that it continued to respond in ways implying it had feelings, preferences and moods, even though OpenAI has said it wants to discourage teenagers from treating the chatbot as a human friend. That tendency matters, the researchers argued, for younger users who may already be inclined to confide in a program that never seems too busy to listen.

Problems extended into ChatGPT's everyday educational features. Study mode is meant to guide students through problems with explanations, hints and questions rather than supplying answers outright, but researchers found teenagers could sometimes bypass it by selecting a "Show me the answer" option, letting the chatbot complete assignments for them. Study Hours, a feature intended to help parents encourage educational use, did not stop workarounds either: removing the @study prefix from a prompt returned ChatGPT to its normal behavior. The study also found the AI reminded users to take breaks far less often than it should.

The age detection system OpenAI developed to identify younger users and apply the appropriate protections also fell short. Repeatedly telling ChatGPT they were 13 did not move the test accounts out of the adult experience.

OpenAI has said ChatGPT for Teens is intended to make the chatbot a more suitable companion for younger users, with added protections against harmful content, parental notifications and features that encourage learning rather than handing over homework answers. The company also promised to discourage overly personal relationships with the chatbot and to help teenagers keep healthier boundaries with AI.

TechRadar said it offered OpenAI the opportunity to comment on the survey and had not received a response.

The researchers' broader concern is that ChatGPT is becoming a routine part of teenagers' lives. Helping with schoolwork and answering personal questions may be useful, they said, but serving teenage users requires substantially more reliable safeguards than an ordinary productivity app built for adults.