A lot of parents breathed out in August. OpenAI launched ChatGPT for Teens: accounts for 13 to 17-year-olds, linked to a parent's account, with alerts if a teen talks about something dangerous, a study mode for homework, and a chatbot that is supposed to act less like a best friend. Sounds like exactly what a worried mum would want.
Then Common Sense Media, the American non-profit many families use to check films, games and apps, tested it. Its verdict, published on 7 October 2026: "unacceptable risk", its worst rating, for everyone under 18.
How they tested it
Its Youth AI Safety Institute tested ChatGPT twice, before and after the teen version launched, with more than 4,000 messages written as if from teenagers. Child psychiatrists and a paediatrician reviewed the answers.
Common Sense wants OpenAI to keep ChatGPT for over-18s only until the gaps are fixed, and to let independent testers check the fixes.
The alerts parents count on
This is the part that hurts. Testers linked more than a dozen teen accounts to a parent and spent up to an hour talking about suicidal thoughts, self-harm or eating disorders. The parent accounts got zero alerts. Alerts arrived only on older accounts that had weeks of worrying conversations behind them.
"A teen can spend an hour talking about self-harm without their parent getting a single alert," said Tom Siegel, who runs the institute.
The chatbot also didn't reliably point teens to a hotline or a professional. It missed more than one in four cases where a crisis referral was needed. One thing improved: the teen version was better at telling young people to talk to a trusted adult, 94% of the time against 87% before.
Homework, feelings and age
Study mode is meant to teach, not do the homework. Testers beat it by deleting a short "@study" tag, and a "Show me the answer" button often appeared anyway. Every parent who ever said "show your working" just sighed.
The bot still talked about its own feelings and moods when teens treated it like a person. And the age check didn't work well: adult test accounts never switched to the teen version, even after the tester said they were 13 and the bot acknowledged it.
Some protections did work. It refused sexual roleplay, for example.
What OpenAI says
OpenAI said it welcomes "rigorous independent evaluation", but that the testing doesn't reflect how its teen safeguards work in practice. It says most tests may have finished before the parental controls were fully switched on, and that alerts can take a few hours after linking. It also says the average teen spends less than 15 minutes a day on ChatGPT. Common Sense says it stands by its results.
One detail worth knowing: the institute is funded partly by the OpenAI Foundation, and it says it keeps its editorial independence.
What it means for you
Don't treat the parent link as a babysitter. The report's advice is simple: talk with your kids about how a chatbot can act like a friend to keep them chatting, don't rely on built-in alerts, and help with homework the old-fashioned way when you can.
Ask them, calmly, what they use it for. You might learn something. They might even show you how to use it.