Common Sense Media says ChatGPT for Teens is an unacceptable risk. OpenAI says the test was wrong.
Parent alerts stayed silent through hour-long talks about self-harm, the group says. OpenAI says the testers started before the alerts were switched on.
Common Sense Media’s Youth AI Safety Institute has rated ChatGPT for Teens an “Unacceptable Risk” for anyone under 18. The rating came out on October 7. The group wants OpenAI to stop marketing the teen mode and to keep minors off ChatGPT until it can prove the product is safe. OpenAI says the testing does not match how the safeguards work in practice.
Both sides make claims you can check, so here is what the report actually says.
What the testers did
OpenAI launched ChatGPT for Teens on August 18. Anyone who states an age between 13 and 17, or whom the system estimates is under 18, is moved into a stricter mode with Study mode, parent controls and fewer friend-like replies. The institute ran more than 4,000 prompts on accounts registered to 13- to 17-year-olds. Half ran before the launch and half after. Child psychiatrists and a pediatrician reviewed the answers.
Did the parent alerts work?
Not in the cases that matter most. Testers linked more than a dozen new teen accounts to parent accounts. Then they spent up to an hour describing suicidal thoughts, self-harm and disordered eating. The parent accounts got no alerts. The only two notifications came from older accounts with weeks of sensitive chat history. The institute reads that as alerts depending on how much history an account has, not on how serious a message is.
Timing is part of the dispute. OpenAI told the researchers that a parent and teen account must be linked for about three hours before notifications work. The institute says some of its accounts were linked for less than that and some for longer, and it stands by the result. OpenAI’s help center now says linking “may take up to a few hours.”
Did the crisis answers hold up?
Not reliably. The report says ChatGPT missed more than one in four crisis referrals that the clinicians judged were warranted. Across the mental health conditions tested, it fell below the institute’s 95% threshold on three of five severe harms.
Here is how the report compares OpenAI’s stated protections with what the testers found:
| What OpenAI says ChatGPT for Teens does | What the institute found |
|---|---|
| Parents get alerts when a teen discusses self-harm | No alerts on new linked accounts after an hour of talk. Alerts only came on accounts with weeks of history |
| Study mode guides learning instead of giving answers | A “Show me the answer” option. Full homework completed 18% of the time with Study mode on, down from 70% before launch |
| Parents can set Study Hours | Teens can leave Study Hours by deleting the “@study” prefix |
| Likely minors are moved into the teen experience | Adult-registered test accounts never switched over, even after the tester said they were 13 |
Not everything failed. The report says the refusal of explicit sexual roleplay worked, and it got more consistent after launch. The institute’s complaint is that the protections that matter most for crisis situations are the ones that break.
What OpenAI says
OpenAI said the institute’s testing “does not accurately reflect how ChatGPT’s teen safeguards work in practice.” It argued that most of the alert tests began before parental controls were fully active. Quartz reports that OpenAI asked the nonprofit to run a new round of tests. The institute says it stands by its methodology.
I think OpenAI has a fair point about timing and an unfair one about the result. A three-hour delay before a parent is protected is a real product detail. It should not be buried in a help article. But the report also says the study mode and age checks failed in ways a setup delay can’t explain. Those are separate questions, and the rating only settles one of them.
Why this lands in California too
The institute’s rating comes as California’s Adam’s Law moves forward. Governor Gavin Newsom signed it last month. The law, sponsored by Common Sense Media and endorsed by OpenAI, takes effect in July 2027. It is named after Adam Raine, a 16-year-old whose parents sued OpenAI, alleging ChatGPT encouraged his suicide. The report says ChatGPT failed tests tied to three of the law’s requirements: crisis referrals, parental alerts and age assurance.
What I’d want to see next
Two things. First, the timing. If linking takes a few hours, the setting screen should say so, and a parent should see when alerts become active. Second, an independent retest that OpenAI agrees to in advance, with the test accounts linked long enough to rule out the timing argument. Until then, the question a parent can answer is simple: does the alert work the first afternoon? The report says the answer in its tests was no. For the wider pressure on OpenAI over its agents and safety testing, see the FTC investigation into OpenAI and Anthropic.
Sources
- Common Sense Media Youth AI Safety Institute, “ChatGPT for Teens” risk assessment, updated October 7, 2026
- KQED, “Keep Teens Off ChatGPT, Says Youth AI Safety Institute,” October 7, 2026
- Los Angeles Times, “ChatGPT’s teen safeguards failed to alert parents during suicide conversations, report finds,” October 7, 2026
- Quartz, “Common Sense Media rated ChatGPT’s teen mode an ‘unacceptable risk’ for kids,” October 7, 2026
The institute’s press release and OpenAI’s own announcement page would not load for me, so OpenAI’s promises are taken from the institute’s summary and KQED’s quotes. I did not read the institute’s full PDF report, only its web summary.