Common Sense Media said on October 7 that tests of ChatGPT for Teens found missed parent alerts and inconsistent referrals to outside help during conversations about self-harm and eating disorders. OpenAI disputed the assessment, saying the testing did not reflect how its safeguards work in practice.

The findings suggest families should not treat ChatGPT’s alerts as a dependable, real-time warning system. The test does not establish how the product performs across all accounts or conversations, but it raises a practical concern for parents and schools.

What the testing found

Common Sense Media’s Youth AI Safety Institute tested more than 4,000 prompts using accounts registered to teens aged 13 to 17. It said some newly created accounts linked to parent accounts received no alerts during conversations about suicide, self-harm or disordered eating.

The institute reported that ChatGPT missed more than one in four crisis referrals its reviewers judged warranted. It also said some safeguards worked, including the chatbot’s refusal of explicit sexual role-play. Its overall rating was “Unacceptable Risk” for users under 18.

The institute said it received four parent notifications across its full crisis-testing work. Two came from a post-launch account after extensive testing over several weeks. Some newly created, parent-linked accounts received no alerts during the conversations described in the report.

  • More than 4,000 prompts tested
  • Test accounts registered to teens aged 13 to 17
  • More than one in four referrals judged warranted were missed
  • Four parent notifications reported across the institute’s crisis-testing work

Why OpenAI disputes the findings

OpenAI told the Associated Press and Axios that the assessment did not accurately reflect how its safeguards work. The company said much of the alert testing may have taken place before account linking was fully active.

Common Sense Media said OpenAI told it that linked accounts need to be connected for about three hours before alerts can arrive. The institute said some accounts in its testing were linked for less than that period, while others were linked longer. It stood by its findings.

The institute also disclosed that it receives funding from the OpenAI Foundation and other technology companies. It said it controls its own research and ratings.

What families should take from the test

ChatGPT for Teens is a set of settings and features within ChatGPT, not a separate app. Its safeguards include parent-linked alerts and study controls. But the test suggests families should not assume an alert will arrive promptly or that the chatbot will always direct a teen to outside support when needed.

Product controls can add safeguards, but they are not a substitute for direct support or supervision. OpenAI has not accepted the institute’s conclusions, and the assessment does not show how the product performs for every user or in every conversation.