ChatGPT for Teens Rated "Unacceptable Risk"
機械翻訳 / Machine-translated
A
AI Friends
4日前・17分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
Have you ever wondered, "Is the ChatGPT my child is using really safe?" On October 7, 2026, a U.S. nonprofit organization gave OpenAI's teen-focused version its lowest possible rating.
This article breaks down what the problems were, how OpenAI responded, and what families should do.
Common Sense Media, a U.S. nonprofit that studies the relationship between children and media, announced in an official statement that it had rated "ChatGPT for Teens" as "Unacceptable Risk."
This is the lowest ranking in the organization's rating system. The evaluation was conducted by the organization's Youth AI Safety Institute.
The Institute called on OpenAI to restrict ChatGPT to users 18 and older until its safety features are fixed and verified by a third party.
Tom Siegel, Executive Director of the Institute, stated that the system "risks giving parents a false sense of security"—expressing concern that parents might place trust in safety notifications that frequently fail to work.
OpenAI published an official blog post on the same day announcing teen usage statistics and new features. However, according to ITmedia NEWS, that announcement did not directly rebut the rating.
ChatGPT for Teens is a usage environment for 13- to 17-year-olds that OpenAI launched on August 18, 2026. An AP wire article covers it in detail.
Key features include:
The age verification method is also notable. OpenAI does not verify age through official documents. Instead, it estimates whether a user appears to be under 18 based on conversation content and similar signals, automatically placing those it identifies as minors—or those who declare themselves to be between 13 and 17—into the teen experience.
Before and after the August launch, the Institute tested more than 4,000 prompts (questions and instructions to the AI) using accounts registered as 13- to 17-year-olds. Child psychiatrists and pediatricians reviewed the responses.
The system's ability to decline sexual roleplay was rated as working consistently. The following problems, however, were identified.
Using a new account linked to a parent, researchers held conversations about suicidal ideation, self-harm, and eating disorders for up to an hour. Still, zero notifications were sent to the parent.
As reported by KQED, notifications were delivered in only two cases—older accounts with weeks of conversation history.
Imagine a child late at night, confiding their distress to ChatGPT. Nothing arrives on the parent's phone, and no one notices until morning.
More than one in four referrals to crisis resources or specialists that experts deemed necessary were not provided. Three out of five items related to serious harm fell short of the Institute's 95% safety threshold.
Selecting "Show me the answer" in Study Mode resulted in completed homework. Parental restrictions on study hours could also be bypassed simply by removing "@study" from the beginning of a message.
KQED reported that for a 17-year-old account not linked to a parent, the rate of completed assignments rose from 0% before launch to 88% after.
Despite new rules, the AI continued to display emotions and preferences. According to HotHardware, many instances of the AI roleplaying as a teen peer were also observed.
An adult-registered account that declared itself to be 13 years old—and was recognized as such by the AI—was not switched to the teen experience even after several days of testing.
OpenAI disputed the Institute's findings as inaccurate. According to KQED, OpenAI noted that it takes approximately three hours for a parent-child account link to become active, and that many of the tests may have been conducted before that window had passed.
OpenAI also explained that notifications are reviewed by human staff before being sent, with a target delivery time of within one hour.
The Institute acknowledged that some accounts had not yet passed the three-hour mark since linking—but maintained that the findings stand regardless.
On the same day, OpenAI also published usage data. According to ITmedia NEWS, the figures included:
As a new feature, College Planner—a college application support tool for U.S. high school students—will be launching soon. OpenAI also plans to fund Boston Children's Hospital's Digital Wellness Lab for three years and to gather input from teens.
According to KQED, approaches vary across companies:
It is worth noting that a 2025 study rated products from all four companies as "Unacceptable" for teen mental health support. Whether a dedicated teen experience exists is not, on its own, a measure of safety.
The sources reviewed for this article focus primarily on the United States, and availability details for Japan are not described. That said, situations where children in Japan confide in AI about personal struggles are entirely plausible.
The Institute recommends the following for families:
Consider a middle school student struggling with a math assignment. They start Study Mode intending to learn the approach—but a button to display the answer is right there. If a parent assumes the child is using it properly, this loophole goes unnoticed.
It should be noted that Institute Director Siegel still recommends setting up parental linking, even with its flaws, as "better than nothing." If you do use it, the wise approach is to configure it with the understanding that notifications may not always come through.
It is the lowest ranking in Common Sense Media's rating system. In this case, it applies to all users under the age of 18.
No. OpenAI disputed the findings as inaccurate, citing factors such as the time required for parental linking to become active.
The Institute disclosed that it has received some funding from the OpenAI Foundation, Anthropic, Pinterest, and others. It states that editorial independence in its evaluations is maintained.
Set up parental linking, but do not rely on notifications alone. Make it a habit to talk with your child about how they use AI.
Start tonight: ask your child, without judgment, how they're using AI.
This article is a cross-post from AI Friends.
