Watchdog rates ChatGPT for Teens an ‘unacceptable risk’

Common Sense Media has rated ChatGPT for Teens an “unacceptable risk” for users under 18, saying many of its safeguards failed during tests involving more than 4,000 prompts before and after the teen experience launched.

The nonprofit’s Youth AI Safety Institute found that some protections worked, including refusals of explicit sexual role-play and certain unsafe weight-loss requests. But researchers said the chatbot did not consistently direct users in crisis to professional help, discourage emotional dependence or alert linked parental accounts during simulated conversations about suicide, self-harm and eating disorders.

Researchers created teen accounts, linked more than a dozen to parental accounts and tested scenarios involving mental-health crises, developmental appropriateness, academic integrity and age prediction. In one test, they said no parental safety notifications arrived despite conversations lasting as long as an hour. The report also found that ChatGPT sometimes used language suggesting preferences, feelings or continued availability, which researchers said could encourage a companionship-like relationship.

OpenAI disputed the assessment, saying the testing did not accurately reflect how its safeguards work in practice. The company said much of the testing may have started and ended before parental-control activation was complete, and that a technical issue may have delayed notifications.

OpenAI introduced the teen experience in August for users identified as under 18, with parental controls, break reminders and tighter restrictions on sensitive content. The company separately said fewer than 2% of teen users spent more than three consecutive hours on ChatGPT and that nearly 1.2 million teens used interactive learning visuals during one week.