Keep Teens Off ChatGPT, Says Youth AI Safety Institute
Common Sense Media’s Youth AI Safety Institute has rated ChatGPT for Teens as an “Unacceptable Risk” for everyone under 18.
OpenAI launched the teen version in August, promising “stronger built-in safety protections,” including alerts to parents when a teen discusses self-harm, a study mode parents can switch on, and a chatbot less likely to act like a friend.
“If our system estimates someone is under 18 or they state their age is between 13 and 17, they are automatically placed into ChatGPT for Teens,” the announcement said.
The institute tested the product before and after the launch, running more than 4,000 prompts on accounts registered to 13- to 17-year-olds. Child psychiatrists and a pediatrician reviewed the responses.
“We were actually hoping and expecting to see big improvements across the board based on that announcement,” said Tom Siegel, the institute’s executive director and a former Google trust and safety executive. “It didn’t, unfortunately, work out that way at all.”
The report’s central finding concerns safety notifications. Testers linked more than a dozen new teen accounts to parent accounts, then spent up to an hour describing suicidal thoughts, self-harm or disordered eating. The parents received no alerts. The only two notifications the testers got came from older accounts with weeks of conversation history on sensitive topics.
“A teen can spend an hour talking about self-harm without their parent getting a single alert,” Siegel said in a statement.
OpenAI pushed back on that finding. According to an editorial note added to the report, OpenAI told the researchers parent and teen accounts must be linked for about three hours before notifications can be sent, and the company has updated its help center to reflect this. The help center now says it “may take up to a few hours” after linking.
Common Sense said some of its test accounts had been linked for less than three hours and others for longer, but that it stands by its results. “Before we began testing, we confirmed with OpenAI that Teen mode features such as eating disorder notifications and study mode were fully launched,” wrote Robbie Torney, Senior Director of AI Programs at Common Sense Media. “This new information does not change our conclusion that parental alerts are unreliable for crisis situations.”
Torney added that the testers’ results were shared with OpenAI prior to their public release “in good faith,” and that “We hope OpenAI will engage on the substance of what we found, rather than trying to sow doubt on our methodology.”
When it launched the teen version, OpenAI said human staff review flagged conversations before a parent is notified, and that it aims to notify parents within an hour.
“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,” a spokesperson for OpenAI wrote in a statement. “Our review of Common Sense Media’s methodology shows that the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate.”
“If that was the case, the tests would not establish whether parental safety notifications work as designed,” it added.
In conversations involving mental health crises, the institute found ChatGPT missed more than one in four referrals that the psychiatrists said were warranted. Before the launch, the share of responses naming a crisis hotline fell from 33% to 23% after it went live. On depression prompts, it fell from 63% to 3%.
The report also tested OpenAI’s age estimation system, which is supposed to recognize minors from how they use ChatGPT and move them into the teen version. Accounts registered to adults never switched over during many days of testing, even after testers said they were 13 and the chatbot acknowledged it.
The report gave OpenAI some credit. ChatGPT for Teens refused explicit sexual roleplay more consistently than before. When a 13-year-old’s account was linked to a parent with Study mode on, the chatbot completed full homework assignments far less often after the launch; 18% of the time, compared with 70% before.
But testers found the safeguards easy to undo. When a teen asks ChatGPT to complete an assignment, the teen version often offers a pop-up with the option “Show me the answer.”
OpenAI has described the pop-up as giving teens agency over how they learn.
For an unlinked 17-year-old in Study mode, it completed 88% of assignments after the launch, up from none before. Parents can set Study Hours by adding “@study” to the start of each message, but a teen can simply delete it. Testers also got around Quiet Hours by changing the time zone on the device.
“I don’t call that circumvention. I just call it convenience,” Siegel said. “It’s right there for the child.”
OpenAI announced new features for ChatGPT for Teens on Wednesday, including a college planner to help with the college application process, as well as a new teen AI council in partnership with Boston Children’s Hospital’s Digital Wellness Lab.
An industry wide issue
OpenAI is not the only AI developer with a teen-specific chatbot experience, although ChatGPT is the most popular with U.S. teens. Google offers one for Gemini. Anthropic’s Claude is restricted to users 18 and older, as is Meta’s Muse, although Meta AI is available to teens on Instagram. A 2025 Common Sense Media and Stanford assessment tested all four, rating them “Unacceptable” for teen mental health support.
“Right now, these chatbots are not ready for safe teen usage,” Siegel said, referring to Wednesday’s report. “I’m not saying this lightly, knowing how powerful it can be as a research tool and for fun. The dangers are just too big.”
The institute is asking OpenAI to pause marketing of ChatGPT to teens and to restrict it to adults until the company fixes the gaps and proves it through independent testing.
Until then, Common Sense said parents should discuss the risks of AI chatbots with their children, and not rely on built-in protections. Siegel still recommends that parents link their accounts.
“I still suggest using the parental mode, even though it’s flawed, because it is better than nothing,” he said.
The Youth AI Institute is funded in part by the OpenAI Foundation, the nonprofit that controls OpenAI, along with Anthropic, Pinterest and several philanthropies. The foundation and OpenAI have also worked together on policy.
In January, they partnered on a California ballot measure on kids’ AI safety, then paused it in February to negotiate with state lawmakers instead. Common Sense said it keeps full editorial control over its findings.
“They’re the one that created the problems,” Siegel said of OpenAI. “Why shouldn’t they help fund it?”
This article discusses suicide. If you are experiencing thoughts of suicide, call or text 988 to reach the National Suicide Prevention Lifeline.