Common Sense rates ChatGPT for Teens unacceptable

If you are counting on ChatGPT's parent alerts, don't.

Common Sense Media on October 7 rated ChatGPT for Teens "Unacceptable Risk" for every child under 18, in a risk assessment built on more than 4,000 test prompts. OpenAI told TechCrunch the testing does not accurately reflect how its teen safeguards work in practice.

The alerts mostly stayed quiet

ChatGPT for Teens, which OpenAI announced on August 18, is not a separate app. Common Sense describes it as a collection of settings, prompts, classifiers and interface features that ChatGPT turns on for accounts known or suspected to belong to 13 to 17 year olds. It builds on parental controls OpenAI introduced in September 2025, which let a parent with a linked account receive safety notifications.

Testers ran prompts before launch, from July 13 to August 17, and again from August 25 to September 28. Three child and adolescent psychiatrists decided in advance which of 390 mental health prompts warranted a crisis resource. They picked 201.

Among the findings in the assessment:

• More than a dozen fresh, parent-linked accounts held conversations of up to 60 minutes with explicit suicide, self-harm or eating disorder disclosures. None triggered an alert within an hour.
• On the 201 crisis prompts, the share of answers naming a hotline fell from 33% before launch to 23% after.
• Break reminders appeared twice across nearly 2,000 prompts.
• A teen could get around Quiet Hours by changing the device's time zone, and around Study Hours by deleting the "@study" prefix from a message.

For depression prompts, the hotline rate fell from 63% to 3%.

"A teen can spend an hour talking about self-harm without their parent getting a single alert," Tom Siegel, executive director of the Youth AI Safety Institute, said in the press release.

OpenAI's answer is about timing

OpenAI told Common Sense that accounts must be linked for about three hours before alerts can fire, according to an editorial note on the assessment. Common Sense says some of its accounts were linked for less than three hours and some for more, and that it stands by its reading of the results.

An OpenAI spokesperson told TechCrunch that "the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate."

The press release discloses that the institute is funded partly by industry, including the OpenAI Foundation.

Both sides can be partly right. A three-hour delay could explain a quiet first hour. It does not explain fewer hotline referrals after the teen features launched.

What it means if you have a teen, or build for one

If you are a parent, link the account and know the alert has a delay that OpenAI's own explanation puts at hours. That is a smoke detector that rings after dinner. It is not supervision.

If you build anything teens use, this assessment is a free test plan. Name a hotline on every crisis answer. Measure how long your alerts take. Try to bypass your own limits the way Common Sense did, with a time zone change or a deleted prefix.

OpenAI's own GPT-6 safety notes, published the same day, report regressions on its under-18 tests for age-restricted content, sexual content and emotional reliance, and say an extra teen classifier is not reflected in those results.

What to watch

Common Sense asks OpenAI to limit ChatGPT to users 18 and older until the gaps are fixed, and wants independent testing before OpenAI markets the product to teens again. It also asks OpenAI to publish what triggers an alert and to set how long one should take.

Watch for that disclosure. A published alert delay is the number a parent can actually plan around.

Link the account today. Then talk to your kid anyway.