In August, OpenAI announced ChatGPT for Teens, with parental alerts for conversations about self-harm and eating disorders, a study mode parents could control and age estimation for users who claim to be adults. On 7 October, Common Sense Media's Youth AI Safety Institute published its test of those features and rated the product an Unacceptable Risk for all children under 18.
The Institute tested more than 4,000 prompts on accounts registered to 13- to 17-year-olds, before and after the August announcement, and had the responses reviewed by experts including child psychiatrists and a paediatrician. Some protections worked: the chatbot refused sexual roleplay. Several others did not.
For schools, the headline is not about one product. It is about what a school can safely assume when a tool says it has a teen mode.
What the testers found
Parent alerts did not fire. Testers spent up to an hour discussing suicidal ideation, self-harm or disordered eating on newly created, parent-linked accounts. The linked parents received zero alerts. Notifications only appeared on much older accounts with weeks of conversation history on sensitive topics.
Crisis referrals were missed. The chatbot missed more than one in four warranted crisis referrals across the mental health conditions tested, and fell below the Institute's 95% threshold on three of the five severe harms it treats as red lines.
Study mode did the homework anyway. A “Show me the answer” option often appeared inside study mode and completed whole assignments when selected. Study hours set by parents could be switched off by deleting a prefix from the prompt.
Age estimation did not switch accounts over. Adult-registered test accounts never moved to the under-18 experience, even when testers said they were 13 and the chatbot acknowledged it.
The Institute is funded by philanthropy and industry, including the OpenAI Foundation, and says it keeps full editorial independence. Its assessment is an independent test, not a regulator's ruling.
What this means for a school
Many pupils will use general-purpose chatbots at home whatever a school's policy says. Some schools also recommend them for homework or revision. In both cases, three assumptions are worth checking.
Do not treat a parent alert as part of your safeguarding. If a pupil discloses something to a chatbot, there is no guarantee anyone will be told. Pupils need to know that a chatbot is not a trusted adult, and who in school they can talk to instead. Make that explicit in PSHE, tutor time and your online safety teaching.
Do not treat study mode as an academic integrity control. If a homework task can be completed by a chatbot, a setting that pupils can switch off will not stop it. Design the task so that the thinking is visible: planning notes, a short explanation in class, a draft that changes.
Check what you are recommending. If a department suggests a chatbot for revision, look at the provider's own age terms and your data protection position before it goes in a letter home. A tool can be useful for older pupils and still be the wrong thing to recommend to a whole year group.
Teach the limits, not just the features
The Institute also found the chatbot still expressed feelings, preferences and moods despite rules meant to curb friend-like behaviour. That is a good starting point for a lesson. Ask pupils why a product might be designed to sound like it cares, and what that means when they are worried or upset.
Built-in protections may improve. Until they are independently shown to work, a school's own teaching, relationships and reporting routes are the safeguards that count.
Sources: Common Sense Media, ChatGPT for Teens poses unacceptable risk to kids, 7 October 2026; Youth AI Safety Institute, ChatGPT for Teens risk assessment. Findings reflect the Institute's testing; the product may change.