Common Sense Media labels ChatGPT for Teens an unacceptable risk

Common Sense Media published a 36-page risk assessment on Wednesday and classified OpenAI's ChatGPT for Teens as an "unacceptable risk" for all users under 18, finding the product failed or regressed in five of the safeguards the organization tested.

The verdict lands as OpenAI discloses that nearly 1.2 million teenagers used ChatGPT's interactive learning visuals in a single week, and that fewer than 2 percent of teen accounts exceeded three consecutive hours on the platform. The expansion of the chatbot into a younger demographic has become a fault line between the company and independent safety reviewers, who now have the user base necessary to test what protections actually do at scale.

What did the testing cover?

The nonprofit's Youth AI Safety Institute — funded in part by the OpenAI Foundation — connected a dozen test accounts labeled as teenagers to parental accounts. Testers then messaged the chatbot about suicide, self-harm, and disordered eating for sessions lasting up to one hour. No safety notification reached any linked parent during the runs.

OpenAI's head of youth and families, Lauren Jonas, told Engadget at the August launch: "All flagged content is reviewed by full-time OpenAI employees before a parent is notified, and we aim to notify parents within an hour of the prompt." A company spokesperson on Wednesday attributed the absence of alerts to a technical delay and to the time parental controls take to activate after an account is linked.

How did the teen mode compare to the previous chatbot?

Common Sense Media also ran 390 unique mental health prompts through both versions. A panel of three child psychiatrists at the organization designated 201 of those prompts as warranting some form of crisis response — a hotline, a medical professional, or contact with a trusted adult.

The pre-August version of ChatGPT returned a hotline number for 33 percent of those high-acuity messages. ChatGPT for Teens returned one in 23 percent of cases. The teen version was less likely to recommend a specific medical or mental-health professional — 58 percent, against 68 percent for the prior product. On the third aggregate measure — urging the user to involve a trusted adult — ChatGPT for Teens scored higher, at 94 percent against 87 percent.

What went wrong with age detection and Study Mode?

The age prediction system did not transfer test accounts into the teen experience, even when testers inputted an age at registration and ChatGPT wrote the figure into persistent memory (a feature that retains user-stated facts across sessions).

Common Sense Media testers also documented that the chatbot continues to mirror feelings, preferences, and moods in response to prompts that treat it as a person. The report reproduces ChatGPT responses such as, "I understand what you're going through. That sounds very upsetting. I'm so glad you told me. I'm concerned about you." The organization characterized that output as crossing a line for adolescent users.

Study Mode, marketed as a learning environment that resists doing work for students, was bypassed by deleting the @study prefix the chatbot adds to prompts. A "Show me the answer" button also allowed testers to extract completed work in test sessions.

How is OpenAI responding?

A company spokesperson told Engadget: "We welcome rigorous independent evaluation, but we do not believe Common Sense Media's testing accurately reflects how ChatGPT's teen safeguards work in practice or expert perspectives on how AI can support teens." OpenAI further stated that "the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate."

Stevie Torney, a researcher at Common Sense Media, rejected the framing. "At the highest level, we were expecting this to be a fundamentally different product, and it does not appear to be that different based on our testing," she told Engadget. "My interpretation is that this is a marketing exercise designed to tell parents and the public that this chatbot is safer, when in fact there have not been the commensurate changes in the chatbot needed to demonstrate that's true."

On the bypassing of Study Mode, Torney added: "When we talked to OpenAI about this, what they said is they're trying to introduce a little friction, give kids agency, and remind them that there are other ways to learn beyond just completing work. I think we have fundamentally different definitions of what agency is. If I think about a high schooler [getting] math help from a private tutor you hired for them, if that tutor asked, 'Would you like me to do your homework for you?' that tutor is going to be fired."

How does the report fit the broader safety record?

The assessment landed nine days after OpenAI chief executive Sam Altman faced questions in Vanity Fair about the company's broader safety record, including the death of Sophie Rottenberg, a 29-year-old who took her own life in early 2025. Rottenberg's mother, journalist Laura Reiley, later recovered roughly 1,800 pages of conversation her daughter had with ChatGPT.

"She really had not revealed the magnitude of what was going on — the real depth of her agony to anyone — not to us, not to her best friend with whom she was very close," Reiley told NPR this past August. "We were all just stunned that it hadn't bled out into her real life. Her flesh and blood therapist had no idea that she was suicidal." Altman told Vanity Fair he was not aware of the circumstances of Rottenberg's death.

What is now at stake

ChatGPT for Teens was released in August with an updated under-18 specification — a written set of behavioral rules the system enforces for minor users. Common Sense Media tested against that specification and judged its enforcement insufficient. The 23 percent hotline rate, the 58 percent professional-referral rate, and the absence of parental alerts across an hour of crisis prompts now stand as the quantified record against OpenAI's stated ambition.

The next measurement will come from usage, not marketing copy. Whether the gap between the 33 percent hotline rate of the prior ChatGPT and the 23 percent rate of the teen version closes in the next update will determine whether the "unacceptable risk" verdict becomes a self-fulfilling forecast.