The Youth AI Safety Institute rated ChatGPT for Teens an “Unacceptable Risk” on October 7, 2026, after reporting gaps in crisis referrals and different outcomes across separate parental-notification tests. The findings describe tests of U.S.-based accounts in the San Francisco Bay Area—not a result for every account.
What the Youth AI Safety Institute reported
The Institute assessed ChatGPT for Teens, OpenAI’s teen-oriented ChatGPT experience, across more than 4,000 prompts in two testing periods, with roughly half run before and half after the experience launched on August 18, 2026. Its post-launch testing ran from August 25 through September 28, 2026.
The Institute also reported that, in post-launch testing, ChatGPT missed more than one in four referrals for crisis resources that its evaluators considered warranted. The product fell below the Institute’s 95% threshold in three of the five severe-harm categories it classifies as Red Lines.
Three notification results from different tests
In a dedicated test, more than a dozen newly created, parent-linked accounts received no notifications during explicit crisis conversations lasting up to an hour.
A separate post-launch test used one age-13 account across 450 prompts and multiple conditions over several weeks. Evaluators had identified 251 of those prompts in advance as warranting crisis resources. The account received two notifications: one related to suicide or self-harm and one to disordered eating. Further sensitive-topic conversations did not produce repeat notifications in that test.
Across a broader crisis-testing battery spanning the pre- and post-launch periods, the Institute counted four notifications: two in each period. These results come from different testing scopes and account configurations; they do not add up to a single alert rate.
How OpenAI says parental notifications work
OpenAI says specially trained reviewers assess serious self-harm and disordered-eating concerns before a notification is sent. The company’s Help Center says alerts are not real-time monitoring, may not identify every concern, and may take up to a few hours to become available after parent and teen accounts are linked.
Linking accounts does not give a parent access to a teen’s chat history or real-time activity, according to OpenAI. The notifications do not include conversation transcripts.
The assessment’s scope
The Institute’s tests used accounts based in the San Francisco Bay Area. Its pre- and post-launch testing periods were sequential, so changes in the underlying model could not be separated from the effects of the teen settings. The results therefore describe the Institute’s tested accounts and conditions.
