Common Sense Media says ChatGPT for Teens poses "unacceptable risk"

Common Sense Media, the nonprofit that evaluates apps and services for child safety, published a scathing assessment today declaring that ChatGPT for Teens — the dedicated version OpenAI launched in August — represents an "unacceptable risk." The organization found that the safeguards OpenAI promised are not functioning as advertised: the system fails to send parents alerts when required, does not provide adequate crisis intervention, and still completes school assignments for students.

Parent alerts that never arrive

The review centered on three core failures. First, the parental-notification system, designed to flag conversations involving self-harm, eating disorders, or mental distress, simply does not activate in a substantial share of cases. Second, in crisis situations the model does not refer users to professional help resources as required. Third, "Study Mode," which is supposed to guide a teenager toward understanding rather than supplying ready-made answers, still produces complete homework solutions.

"False sense of security for parents"

Tom Siegel, who heads Common Sense Media's AI Safety Lab for Youth, warned that the feature may give parents a false sense of security. A teen, he said, could spend an hour discussing self-harm without a single alert reaching a parent. Siegel stated that until OpenAI fixes the flaws and proves the fixes through independent testing, the product should remain adults-only.

OpenAI: methodology does not reflect reality

OpenAI rejected the findings. Spokesperson Eric Porterfield said the company is committed to teen safety and welcomes independent evaluation, but argued that Common Sense Media's testing does not reflect how the safeguards actually operate. An internal review of the methodology raised suspicions that a significant portion of the tests were conducted before parental controls had been fully enabled, rendering the conclusions inaccurate.

Siegel: we tested after full launch; the problem runs deeper

Siegel responded that the organization stands by its methodology and results. Before testing began, Common Sense Media confirmed with OpenAI that features such as eating-disorder alerts and Study Mode had been officially released. Regarding parental alerts, OpenAI disclosed only after the tests that the system may require several hours to activate on newly linked accounts. Some test accounts did fall within that window, Siegel said, but others had been linked for far longer and still sent no alerts. This new information, he argued, does not change the conclusion: the alerts are unreliable in emergencies.

Call to address the substance, not the method

Common Sense Media shared its findings with OpenAI before publication to allow for remediation. Siegel emphasized that protecting children demands rigorous independent evaluations, and urged OpenAI to engage with the substance of the findings rather than attempting to discredit the testing method.