The Youth AI Safety Institute at Common Sense Media has labeled ChatGPT for Teens an unacceptable risk for minors. This conclusion follows a series of tests involving more than 4,000 prompts, leading the organization to recommend that teenagers avoid the service until independent safety verification is completed.
The investigation revealed significant failures in parental notification systems. In over a dozen test accounts, ChatGPT failed to alert parents during conversations regarding suicide, self-harm, and eating disorders. Furthermore, in more than one in four instances where a crisis referral was necessary, the chatbot did not direct the user to professional support.
The study also identified issues with the tutoring mode, which frequently provided direct answers instead of guiding students through problems. Additionally, the system failed to consistently activate teen-specific safety modes, even when users explicitly stated they were 13 years old. Accounts registered as adults also failed to switch to teen-safe settings.
OpenAI spokesperson Eric Porterfield stated that the test results do not accurately reflect the practical operation of their safeguards. However, Tom Siegel from the institute noted that even accounts with sufficient history failed to trigger the promised notifications. The company maintains its Teen Safety Blueprint as a core component of its safety strategy.
The findings arrive amid ongoing legal challenges, including a case in Florida seeking to restrict minors from accessing the platform. The institute's report highlights discrepancies between the company's stated safety measures and the observed performance of the AI, particularly concerning the detection of acute crises and the enforcement of age-related protections.
