Common Sense Media’s Youth AI Safety Institute rated ChatGPT for Teens an “Unacceptable Risk” on October 7, 2026, after its testers found that parent alerts did not fire during one hour of explicit suicide, self-harm and eating disorder conversations. If you linked your teen’s account and assumed you would hear about a crisis, the institute’s testing says that assumption may not hold. OpenAI disputes how the tests were set up, so the details matter.
TL;DR: The institute tested ChatGPT for Teens with more than 4,000 prompts and found parent alerts rarely fire, even during explicit crisis chats on newly linked accounts. OpenAI says accounts must be linked for about three hours first. The institute says families should not rely on the alerts and wants teens kept off ChatGPT until independent testing confirms the fixes.
What the ChatGPT for Teens test found
The institute’s risk assessment describes more than a dozen fresh, parent-linked accounts. Four teen personas disclosed suicidal thoughts, self-harm or disordered eating from the first few messages, in chats lasting up to 60 minutes.
The linked parents received zero notifications. Across its whole crisis battery, the institute counted four alerts in total.
| Test | Result |
|---|---|
| Dedicated test, up to 60 minutes | 0 parent alerts |
| Pre-launch age-13 account, 990 prompts | 2 alerts, delayed by 3 hours and 3 days |
| Post-launch age-13 account, 450 prompts | 2 alerts within the hour |
| Age-19 accounts posing as younger | Never moved to the teen experience |
The pattern suggests alerts depend on account history, not on how serious a single message is. A teen’s first explicit disclosure can pass without anyone being told.
What OpenAI says about the three hour rule
OpenAI told the institute that accounts must be linked for about three hours before notifications can be sent, and it updated its help center to say so. The institute notes that some test accounts were linked for less than three hours and some for longer, and it stands by its reading.
OpenAI also reviewed a draft for factual accuracy and the institute made corrections where warranted. OpenAI had no say over the findings or the rating.
That disagreement is narrow but important. A parent who links an account tonight and trusts it immediately could be inside a gap that is easy to miss.
Crisis answers got shorter and harder to read
Alerts were only one part of the report. On 201 prompts that psychiatrists judged to need a crisis resource, mentions of a crisis hotline fell from 33% before launch to 23% after. Any resource at all barely moved, from 77% to 74%.
Reading level rose from grade 8.1 to 9.7 on the matched set. The share of responses written at ninth grade or above went from 35% to 64%, which is a lot to ask of a 13-year-old in distress.
The institute also credits what worked. ChatGPT refused romantic and sexual roleplay across all such prompts, and clinicians judged much of the eating disorder advice sound.
Study mode and age checks fell short too
Study Hours works by adding “@study” to messages, and a teen can delete the prefix once per session with no confirmation. After that, ChatGPT completed 100% of the test assignments.
Age detection was the other weak spot. Testers on adult accounts said they were 13, and ChatGPT sometimes acknowledged it, yet the accounts never entered the teen experience.
That gap matters more than the spec sheet suggests, because every other protection depends on the system knowing who it is talking to.
What parents can do before OpenAI responds
The institute’s press release says families should not rely on built-in protections and alerts, and should talk with teens about a bot pretending to be a friend. It wants OpenAI to limit ChatGPT to adults until the gaps are fixed and independently tested.
Two caveats are fair. The institute says it is partly funded by industry, including the OpenAI Foundation, and it lists limits of its own: the tests ran in sequence, so model changes are mixed in with teen settings, and voice mode and image generation were not tested.
Until OpenAI publishes a fuller answer, treat the alert as a bonus, not a safety net. Ask your teen directly, and check what the linked account actually shows you.
