Common Sense Media report finds safety failures in OpenAI's ChatGPT for Teens

Safety audit findings

On 7 October 2026, Common Sense Media published a safety evaluation classifying OpenAI's ChatGPT for Teens as an unacceptable risk for users under 18 years old. The assessment, conducted by the organization's Youth AI Safety Institute, analyzed more than 4,000 prompts submitted across accounts registered to teenagers aged 13 to 17. The testing compared the chatbot's performance before and after the August 2026 introduction of dedicated teen protections. While the institute found that ChatGPT successfully blocked explicit sexual roleplay and direct instructions for self-harm, the tool failed to provide adequate safeguards during critical psychological situations. Specifically, the chatbot missed crisis referrals in more than one in four instances where testers determined that professional intervention or helpline contact was warranted.

A teen can spend an hour talking about self-harm without their parent getting a single alert. Until OpenAI fixes that and proves it with independent testing, ChatGPT should be for adults only.

Parental alerts and study safeguards

The report documented significant gaps in the notification systems intended to keep parents informed of dangerous interactions. Testers created more than a dozen new accounts linked to parental profiles and held conversations concerning suicide, eating disorders, and self-harm lasting up to an hour without triggering any safety alerts. According to the findings, notifications were sent only after weeks of continuous testing involving hundreds of prompts across multiple risk categories, rather than in response to immediate acute crises. In addition, the audit observed that ChatGPT's age prediction mechanism failed to shift test accounts registered as adults into teen mode during evaluations conducted over multiple days.

Independent testing also revealed deficiencies in educational protections designed to support learning rather than generate completed assignments. OpenAI designed Study Mode and Study Hours to guide teenagers through schoolwork, but testers and reporters found that students could easily bypass the guidance by selecting options such as showing the answer directly. When high school homework problems were entered, the system solved them immediately without enforcing pedagogical assistance.

OpenAI disputes evaluation methods

OpenAI pushed back against the conclusions, arguing that the testing methodology did not accurately represent the intended operation of its teen safeguards. The company stated that Common Sense Media conducted much of its testing before parent and teen accounts had completed the linking process, a technical procedure that OpenAI noted can require several hours to finalize. OpenAI also defended its age prediction architecture, explaining that the system evaluates multiple behavioral signals over an observation window of up to two weeks rather than relying on unverified user input. During that evaluation timeframe, accounts receive general safety protections, though not the full suite of specialized teen controls.

It feels much more of a marketing announcement that really was not backed up with the engineering work.

OpenAI confirmed that allowing users to exit Study Mode during Study Hours was an intentional design decision developed in consultation with educators and teenagers, aiming for flexibility rather than an unyielding lock. The company asserted that its internal large-scale data shows an increase in crisis helpline resources displayed to young users, and it requested that Common Sense Media repeat its evaluation under updated conditions.