Common Sense Media rated ChatGPT for Teens an Unacceptable Risk. It found parent alerts for self-harm and eating disorders often failed. Crisis referrals missed over one in four warranted cases. Study mode still allowed direct answers and bypassing time limits. The bot still mimicked friend-like behavior. Age estimation did not switch adult accounts to teen mode. The Institute calls for a pause on teen access until fixes are verified. It demands independent testing of age checks, alerts, crisis responses, homework limits and conversational tone. OpenAI’s teen protections fell short of its own commitments.
What changed
OpenAI promised stronger built-in safety protections for teens.
What it unlocks
Parents can no longer rely on promised safety alerts and learning protections for teen users.
- 4,000 prompts tested
- 95% threshold missed on 3 of 5 severe harms
- 25%+ crisis referrals missed
Sources