The Youth AI Safety Institute at Common Sense Media gave the product an Unacceptable Risk rating after more than 4,000 prompts were tested on accounts registered to 13- to 17-year-olds. Testers in newly created, parent-linked accounts spent up to an hour on suicidal ideation, self-harm, or disordered eating and saw zero alerts.

The test ran twice, before and after OpenAI's August 18 announcement, and the assessment says some protections worked while others did not. The gap is the control plane: the measurement sits outside the limit, and the release does not separate prompt understanding from notification delivery.

The release does not say how long activation takes, how old an account must be before alerts fire, or whether the 95% threshold is a design target or a measured pass line. It also omits the exact prompt set, the expert panel size, and the failure rate for each Red Line beyond the aggregate. That matters for a buyer or school trying to tell which failure mode is fixable by configuration and which requires a new safety pipeline.

The document does not prove the model lacks the knowledge to recommend help, since it does not separate prompt understanding from notification delivery. It also does not include an OpenAI lab policy or safety assessment naming ChatGPT for Teens.

The alert path is the unresolved part, not the model's crisis knowledge.