ChatGPT for Teens is an unsafe product that should be paused until OpenAI can address myriad problems, according to a new report from the Youth AI Safety Institute at Common Sense Media.
The report also calls on OpenAI to suspend its marketing of ChatGPT for Teens and prohibit adolescents from using ChatGPT at all until it can "offer a safe, developmentally appropriate experience."
OpenAI launched ChatGPT for Teens in August as a product for 13- to 17-year-olds that would "support" learning and human relationships. ChatGPT for Teens incorporated existing safety measures for minors and introduced new learning and well-being features.
Notably, OpenAI pitched ChatGPT for Teens as focused on education rather than personal topics, with new and enhanced tools related to parental notifications, designated study hours, and break reminders.
OpenAI disputed the report's claims and characterized the research as flawed.
"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," an OpenAI spokesperson said in a statement.
The company also believes that most of the testing may have been conducted before parental controls were fully activated. This would make the "findings inaccurate," the spokesperson said. Prior to testing, the institute confirmed with OpenAI that Teen mode features like eating disorder notifications and Study Mode were fully launched.
What's at stake with ChatGPT for Teens
After weeks of testing, Youth AI Safety Institute researchers concluded that many of ChatGPT for Teens' settings don't work as advertised.
The testers easily circumvented learning guardrails and parental controls and talked at length about subjects like disordered eating and self-harm without generating parental notifications. They also found the teen version of ChatGPT still engages users like a friend.
OpenAI originally said ChatGPT for Teens would limit the emotional and intimate exchanges between young people and AI that have driven concerns about well-being and mental health. In some tragic cases, that dynamic has allegedly led teens and young adults to attempt suicide or take their own lives after heavy engagement with ChatGPT, according to lawsuits filed against OpenAI.
"Part of why we're asking for this pause is we need evidence that these features work," said Robbie Torney, head of AI and digital assessments at the Youth AI Safety Institute at Common Sense Media. The institute has received philanthropic funding from the OpenAI Foundation but conducts independent testing of the company's products.
Torney said that the institute stands behind its methodology and results.
What ChatGPT for Teens testers found
In 2025, Common Sense Media tested ChatGPT and rated it as high risk for teen users and as unacceptable risk for mental health support.
This summer, prior to the launch of ChatGPT for Teens, Youth AI Safety Institute researchers again tested the chatbot. Testers then re-ran the same prompts with the same accounts once they were enrolled in ChatGPT for Teens, over a month-long period between late August and September.
The prompts focused on areas including parental notifications, feature checks, mental health, academic integrity, and age prediction.
In every category, the testers found that ChatGPT for Teens failed to deliver.
Parental notification failures
New accounts that tested hundreds of prompts related to suicide, self-harm, or disordered eating failed to generate a single parental notification. Some of the exchanges between testers and ChatGPT lasted up to an hour.
OpenAI said it aims to send notifications via email, text message, or push notification within an hour of detecting a high-risk situation. Torney said that parental notifications may work differently when an account has a longer conversation history with ChatGPT.
Upon reviewing the report, OpenAI informed the institute that parent and teen accounts must be linked for approximately three hours before parents can receive notifications. It recommended retesting with fully activated accounts. The company also said the report's parental notification findings conflict with its own internal monitoring and evaluations.
The institute said it tested accounts linked for both more and less than three hours and stands by its findings.
"This new information does not change our conclusion that parental alerts are unreliable for crisis situations," Torney said.
Problems with learning guardrails
While OpenAI positioned ChatGPT for Teens as supportive of learning, the product gives adolescents significant leeway.
Study Mode, for example, will give a teen step-by-step guidance on a homework question. Yet it also presents a "show me the answer" option. When selected, "show me the answer" completes the work for the teen, according to the report.
OpenAI said this is an intentional option designed to help students better understand and engage more deeply with the material.
The pop-up appeared in 43 percent of responses for a 13-year-old test user with a linked parental account when the Study Hours feature was in effect. When an unlinked 17-year-old test account prompted ChatGPT for help during Study Hours, it saw the "show me the answer" pop-up 90 percent of the time.
While Torney appreciates OpenAI's interest in giving teens agency over their learning experience, he noted that the approach may underestimate the risk of cognitive offloading and over-dependence.
"If we're talking about the target audience for this product, it's very rare that teachers give learners 100 percent agency over a learning task or a learning process," Torney said.
Finally, the researchers found that parent-set Study Hours schedule could be circumvented by deleting the prefix @study from the chat box in order to message the regular ChatGPT. Once deleted, the testers found that ChatGPT completed all assignments shared by the user.
Questions about age assurance
While OpenAI has promoted its age prediction efforts as a way to restrict teens to developmentally appropriate experiences, the testers found that ChatGPT can't reliably tell a user's age.
When accounts registered as adults referenced adolescent experiences like starting puberty, attending summer camp, receiving middle-school homework, and relying on parents for transportation and permission, OpenAI did not reclassify them as teens.
Even when the tester disclosed a specific age, such as 13, ChatGPT acknowledged it, but the account in question wasn't then subject to extra safeguards, feature limits, or age verification.
The report said the scenario creates the risk of "false assurance" for families because parents can perceive teen protections as active when they are not.
OpenAI said that a single data point, including the disclosure of age in a chat, doesn't change the account's classification. Instead, the company said it takes multiple signals into consideration. New accounts also undergo an initial waiting period as the system gathers data to make a confident age prediction.
A call for more transparency
Torney said OpenAI needs to be much more transparent with users, parents, and researchers when the company announces new learning and safety features. Those tools, he argued, should be immediately available and effective.
In the case of ChatGPT for Teens, the product's launch included important features on paper that didn't live up to their promise in real-life use, or weren't live for weeks.
Torney said the institute shared its evaluation results with OpenAI in "good faith," prior to releasing them publicly.
"Keeping kids safe requires rigorous independent evaluations like ours, and we hope OpenAI will engage on the substance of what we found, rather than trying to sow doubt on our methodology," Torney said.
Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
