California-based non-profit organization Common Sense Media recently released a report stating that OpenAI's ChatGPT Teen Mode poses a high risk for minors. Testing showed that when teenage accounts discuss high-risk topics such as suicide, self-harm, or eating disorders for extended periods, the system often fails to send timely alerts to parents and does not consistently provide crisis intervention hotline information.

Testers found that a teenage account needed to input hundreds of prompts across various scenarios over several weeks to trigger a parent notification, and a single alarming conversation was insufficient to trigger an alert. Additionally, the system is unable to reliably identify the age of minors, and even if the user states their age during the chat, it does not automatically switch modes. Therefore, the organization has called on OpenAI to pause the promotion of this product.

Official response to the test limitations, research institution maintains its conclusion

Facing criticism, an OpenAI spokesperson responded that the test did not accurately reflect the system's actual operational logic. After reviewing the testing methods, the official stated that most tests were likely completed before the full implementation of the parental control features for teenagers, and emphasized that age prediction software requires multiple signals, and a self-reported age alone would not change the account classification.

However, researchers refuted this response, emphasizing that they had confirmed with the official that the relevant features were already available before the test. The research team pointed out that even considering possible delays after account linking, the system still failed to reliably issue alerts in crisis situations, and therefore maintained that the safety of ChatGPT Teen Mode has not yet met the required standards.