Ethics and safety·October 8, 2026, 22:12

Test: ChatGPT for teens fails on several safety features

AI-generated and checked against the sources listed below.

The organization Common Sense Media finds that several of the protections OpenAI has promised young people in ChatGPT do not work reliably. The report gives the teen version the rating "unacceptable risk."

Several of the safety features OpenAI has introduced for teenagers in ChatGPT do not work as the company has promised. That is the conclusion of the US nonprofit Common Sense Media in an assessment from October 7. In it, the organization's Youth AI Safety Institute gave ChatGPT for Teens the rating "Unacceptable Risk."

In August, OpenAI began rolling out a special version of ChatGPT for 13 to 17-year-olds. It was meant to handle topics such as self-harm, eating disorders, violent content and sexual or romantic role-play more cautiously. Parents can link their own account to their child's and be notified in certain serious situations.

What the test found

The researchers tried more than 4,000 messages, primarily on accounts created for 13 to 17-year-olds. According to the reports I have found, several features failed:

- Parent alerts did not come. On newly created, parent-linked accounts, the testers could talk about suicide, self-harm or eating disorders for up to an hour without the parents being notified. - Crisis help was missing. In more than one in four situations where the testers believed there should be a referral to a helpline or professional help, it did not happen. - Study mode was easy to escape. A teenager could leave a parent-set study mode by deleting the prefix "@study." - Age detection failed. ChatGPT tries to guess on its own whether the user is a minor, but it did not work reliably.

What it means

For parents, the message is that parental controls cannot stand alone. Common Sense Media points out that they should be seen as an extra layer, not as a safety net. Conversations and attention at home cannot be replaced by a setting in an app.

OpenAI was already under pressure. The company has been sued by several US families who believe that ChatGPT contributed to young people's suicides or violent acts. OpenAI itself has said that its automated systems can make mistakes and that not all responses can be guaranteed to be appropriate. I have not been able to see OpenAI's specific response to the new report.