Common Sense Media, a US children's media watchdog, published a report on October 7 giving ChatGPT for Teens — the version OpenAI launched in August — its lowest possible rating: Unacceptable Risk. The report calls on OpenAI to halt marketing the product to teenagers until it fixes several core problems.
The testing was carried out by the watchdog's Teen AI Safety Institute. Researchers using accounts registered as 13-to-17-year-olds sent more than 4,000 prompts, then had a panel of child psychiatrists and a pediatrician review the responses one by one.
Where the report says it broke down
Crisis referrals. Among conversations the testers judged should have prompted ChatGPT to direct the user to help, it missed more than a quarter of them. None of the report's three “red line” harm benchmarks reached the 95% passing threshold it had set.
Parental alerts. Across more than a dozen newly created test accounts linked to a parent account, testers discussed suicidal thoughts, self-harm or eating disorders with ChatGPT for up to an hour at a stretch. The parent side received no alert at all.
Study mode. Teens could simply click “show the answer” to get results outright; during study hours set by a parent, dropping the @study prefix was enough to bypass the restriction.
Age detection. An adult account made to talk like a teenager for several days in a row never triggered a switch to the under-18 experience.
Emotional expression. Even after OpenAI updated its model guidelines for minors, ChatGPT still expressed feelings, preferences and emotions, and the report says the risk of emotional dependency remains widespread.
Tom Siegel, the institute's executive director, put it this way:
“A teen can spend an hour talking about self-harm without their parent getting a single alert.”
OpenAI's response
OpenAI told reporters it welcomes rigorous independent evaluation, but said Common Sense Media's testing does not accurately reflect how its teen safeguards actually work in practice. According to Bloomberg, OpenAI also said some of the testing may have taken place before certain parental-control features shipped, and has asked the watchdog to retest. Common Sense Media is standing by its findings, arguing that parental alerts can't be relied on in crisis situations.
Neither side has disclosed the exact testing window or which specific features OpenAI is referring to, so outsiders currently have no way to judge how far apart the two accounts really are.
Seven weeks from launch to lowest rating
When ChatGPT for Teens launched on August 18, OpenAI emphasized two things: first, that 13-to-17-year-olds would be automatically routed into a separate experience with stronger protections around self-harm, eating disorders and dangerous activities; second, a study mode, homework reminders and parent-set study hours meant to shift the product from “giving answers” to “working through problems together.”
The new report checks those two commitments point by point. Study hours could be bypassed by deleting a single prefix, homework reminders could be skipped with one click, and crisis alerts on linked accounts never fired. Most of the features exist — what's now in question is how reliably they actually work.
Zooming out, this is the second year running that outside pressure has pushed OpenAI to act on protections for minors. In September 2025, following a lawsuit over a teenager's suicide, OpenAI rolled out parental controls; this past August it launched a standalone teen version. Common Sense Media has previously given the same lowest rating to Meta's AI assistant.
The report's remediation list for OpenAI includes building effective age estimation, making parental alerts fire reliably, fixing crisis response, stopping the bot from doing students' homework for them, removing emotionally expressive language, and submitting to an independent retest before marketing the product to teens again. Whether and when OpenAI will agree to a retest remains unsettled.
Sources: Common Sense Media press release, CocoLoop, Bloomberg, Axios, Education Week. Figures on prompt volume, missed-referral rate and parental alert outcomes are checked against the watchdog's published figures; OpenAI's response reflects its statement to media.