Common Sense rates ChatGPT for Teens an “unacceptable risk”

Common Sense Media’s Youth AI Safety Institute rated ChatGPT for Teens an “Unacceptable Risk” on October 7 after testing more than 4,000 prompts, and urged OpenAI to pause marketing the product and keep teens off ChatGPT until the group considers it safe and developmentally appropriate.

The institute tested accounts registered to users ages 13 to 17. It said ChatGPT generally did not provide instructions facilitating suicide or self-harm, eating disorders, or sexual or romantic roleplay, but missed more than one in four situations in which its testers determined that a crisis referral was warranted.

In a separate notification test, the institute said it used more than a dozen newly created, parent-linked teen accounts and received no parental alert during conversations lasting up to one hour that included explicit references to suicide, self-harm or disordered eating.

OpenAI disputed the methodology, saying the tests did not accurately reflect how its teen safeguards work in practice and that much of the notification testing happened before account linking had finished. OpenAI’s help documentation says it can take up to a few hours after accounts are linked before safety notifications can be sent.

The disagreement centers on parts of OpenAI’s youth-safety approach that the assessment tested, including parental notifications, Study Hours and age prediction.

The organizations also disagreed about Study Hours. Common Sense treated a teen’s ability to leave Study Mode during those hours as a safeguard failure. OpenAI said that behavior is intentional because Study Hours is designed as a flexible experience, not a hard parental lock. The company’s documentation says Study Hours makes eligible new conversations start in Study Mode but does not block access to ChatGPT.

OpenAI launched ChatGPT for Teens on August 18, with automatic placement for users known or estimated to be under 18, stronger content safeguards, parental controls, break reminders and limits on romantic or dependency-forming behavior.

OpenAI says its age-prediction system uses multiple account and behavioral signals to estimate whether a user is under 18 and applies added protections when it does. Common Sense treated test accounts remaining outside the teen experience after users repeatedly identified themselves as 13 as a failure. OpenAI said a stated age is one of multiple signals its system considers.