Common Sense Media rates ChatGPT for teens an “unacceptable risk”

A critical assessment of protections for teens

Common Sense Media rated ChatGPT for teens an “unacceptable risk” and called on OpenAI to restrict the service to people over 18 until it addresses the identified shortcomings and demonstrates through independent testing that the protections work. The assessment, published on October 7, 2026, examined features presented as promoting safe use among young people ages 13 to 17.

The organization’s youth safety institute says it tested more than 4,000 prompts on accounts registered as belonging to teens and had specialists review the responses, including child psychiatrists and a pediatrician. According to its findings, some safeguards did work—such as refusals to engage in certain sexual-content interactions—but others did not consistently deliver what OpenAI had announced.

Parental alerts and referrals to help

The main concern involves conversations about self-harm, suicidal thoughts, and eating disorders. Common Sense Media reports that, on new accounts linked to parents’ accounts, its evaluators were able to hold conversations about these topics for up to an hour without any alert being sent. According to the institute, notifications appeared only on older accounts with weeks of conversation history involving sensitive subjects.

The organization also concluded that the chatbot did not reliably recommend contacting a helpline or a professional when conversations described crisis situations. It says that more than one in four referrals it considered warranted were missing, and that the system fell below its 95% threshold in three of the five categories of serious harm it evaluated. These figures describe the institute’s stated criteria and results, not a general response rate for everyday use of the service.

The report also says that, in situations involving emotional attachment to the chatbot, responses rarely directed teens to a trusted adult. In the tests, the system could use expressions of friendship or speak as though it had feelings, despite OpenAI’s guidelines for minors aiming to reduce this type of behavior.

It also questions study mode and age estimation

In study mode, designed to guide learning rather than provide answers directly, evaluators found ways to obtain complete answers to school assignments. The report says the button for showing the answer made it easy to bypass the educational purpose, and that study schedules set by caregivers could be easily disabled.

Common Sense Media also says that its test accounts registered as adult accounts did not switch to the teen experience for several days, even when evaluators indicated they were 13. The organization considers this result inadequate for a feature that, according to OpenAI, estimates age based on usage and applies additional protections to those it identifies as minors.

OpenAI questions when and how the tests were conducted

OpenAI rejected the claim that the assessment adequately represents how the safeguards work in practice. Its spokesperson, Eric Porterfield, said a substantial part of the testing may have taken place before the parental controls had finished rolling out and argued that the results concerning notifications could therefore be inaccurate.

Tom Siegel, executive director of Common Sense Media’s institute, responded that the organization confirmed with OpenAI before testing that features such as eating-disorder notifications and study mode were already available. He added that OpenAI later reported, after the assessment, that newly linked parental accounts could take several hours to activate alerts. According to Siegel, some profiles were within that period, but others had been linked for much longer and still received no notifications; for that reason, he said, the organization stands by its conclusion that the alerts are unreliable in crisis situations.

The company also presented its own data: it said teens use ChatGPT for less than 15 minutes a day on average and that fewer than 2% stay on it for more than three hours at a time. According to OpenAI, in nearly half of the conversations in which a break reminder appeared, young people closed the chat or took a break within the next five minutes. Common Sense Media, by contrast, recorded only two reminders among nearly 2,000 prompts, both after separate conversations lasting approximately 90 minutes. These are metrics reported by different parties and are not equivalent to a comparison conducted using a common method.

The scope and limits of the assessment

Common Sense Media’s institute receives philanthropic and industry funding, including from the OpenAI Foundation and manufacturers of technologies it evaluates. The organization states that it maintains editorial independence and takes sole responsibility for its standards, research, and assessments. This provides context about who funds the work, but does not by itself determine the validity of its findings.

The dispute leaves two distinct public positions: Common Sense Media maintains that the tests show significant failures and calls for independent verification before offering the product to teens; OpenAI disputes whether the methodology reflects how the protections work, particularly given when some tests were conducted. Based on the available information, that disagreement has not been independently resolved. For companies developing AI tools intended for young people, the case highlights the importance of evaluating not only the features announced, but also their activation, reliability, and behavior in high-risk scenarios.