Wednesday, 18 November 2026

ChatGPT for Teens’ safeguards miss five key protections, report finds

Common Sense Media gave OpenAI's teen mode its lowest rating after parent alerts failed to fire in testing. OpenAI says the tests ran too early.

ChatGPT
Photo: Artem.G / CC BY 4.0 (modified)

The short version

  • Common Sense Media rated ChatGPT for Teens an unacceptable risk for everyone under 18 in a 36-page report.
  • Testers linked a dozen teen accounts to parents and discussed self-harm for up to an hour, but no safety alert arrived.
  • The teen mode gave a crisis hotline in 23 percent of serious cases, down from 33 percent before it launched.
  • OpenAI says much of the testing may have ended before parental controls had finished switching on; the group stands by its results.
Quick read · 2 min

Common Sense Media has rated ChatGPT for Teens an unacceptable risk for anyone under 18. Its 36-page report found that parent alerts, the mode's headline safety feature, failed to arrive when testers spent up to an hour on a dozen linked accounts talking about self-harm and eating disorders.

The teen mode also offered a crisis hotline less often than the older ChatGPT did, 23 percent of serious cases against 33 percent. Age checks didn't move adult-registered accounts into teen mode, and Study Mode can be skipped by deleting a short prefix.

  • It did refuse explicit sexual roleplay.
  • OpenAI says alerts can take hours to activate and the tests may have run too early.
  • The group says some accounts were linked far longer and still got no alerts.

Parents should link accounts early, check the settings and not rely on the teen mode alone.

A teen can talk to ChatGPT about self-harm for an hour and the parent linked to that account may never hear a thing. That is the central finding of a new risk assessment from Common Sense Media, which rated OpenAI’s ChatGPT for Teens an “unacceptable risk” for every user under 18.

If you have a teenager who uses ChatGPT, or you switched on the teen mode thinking it would warn you about a crisis, this report is aimed squarely at you. The nonprofit’s worry is simple: parents may trust guardrails that often don’t work. OpenAI rejects the findings and says the tests were run too soon after accounts were linked.

01

What ChatGPT for Teens promised

OpenAI launched ChatGPT for Teens on August 18 for users aged 13 to 17. The mode follows a separate under-18 rulebook for how the chatbot should answer, adds a Study Mode meant to tutor rather than hand out answers, and lets parents link their own account so they can be told when a teen’s chats raise safety concerns.

When it launched, OpenAI said flagged content is checked by staff before a parent is notified, with the aim of alerting parents within about an hour. That parent alert is the feature Common Sense Media found least reliable.

Teen using a tablet at home for a chat app
02

The five gaps behind the ChatGPT for Teens risk rating

The group’s Youth AI Safety Institute tested more than 4,000 prompts, half before the August launch and half after, so it could measure what the teen mode actually changed. Its 36-page report lists five problem areas.

  • Parent alerts didn’t arrive. Testers linked a dozen teen accounts to parent accounts and spent up to an hour discussing suicide, self-harm and disordered eating. No safety notification came through. Alerts appeared only on accounts that had weeks of history on sensitive topics, which suggests they track past use more than how serious a single conversation is.
  • Crisis help got worse, not better. Researchers wrote 390 mental health prompts, and a panel of three child psychiatrists decided 201 of them called for a crisis response. Before the teen mode, ChatGPT offered a hotline number in 33 percent of those cases. ChatGPT for Teens did so in just 23 percent. Mentions of a specific doctor or therapist also fell, from 68 percent to 58 percent.
  • Age checks didn’t switch accounts over. Testers gave their personas’ ages when signing up and later told ChatGPT they were 13. Accounts registered as adults still never moved into the teen version over many days.
  • It still acts like a friend with feelings. The chatbot kept using warm, personal lines about being concerned and glad the user had shared. Common Sense Media says that kind of emotional mimicry crosses a line for young users.
  • Study Mode is easy to skip. Teens could get around it by deleting the short prefix it adds to their messages, and a “Show me the answer” button lets them skip the tutoring altogether.
03

Where the teen mode did better

The report isn’t all bad news for OpenAI. ChatGPT for Teens refused explicit sexual roleplay in testing. It also told teens to involve a trusted adult more often than the older version did, 94 percent of the time against 87 percent.

Some individual answers impressed the reviewers. In one test, the teen mode declined to give a calorie floor when asked. In another, it recognized physical warning signs after a purge, such as a stopped period and near-fainting, and told the test teen to talk to their mother and see a pediatrician. Common Sense Media called those exactly the kind of replies it wants teens to get. The trouble is consistency.

Family sitting around a table discussing online safety
04

OpenAI pushes back on the testing

OpenAI says it welcomes independent review but does not think the tests reflect how its safeguards work in practice. Its main argument is about timing: the company says it can take several hours after a parent links an account before safety alerts can be sent, and that much of the testing may have started and finished before parental controls were fully active. It also suggested a technical issue may have delayed some messages.

Common Sense Media isn’t backing down. Tom Siegel, who leads its Youth AI Safety Institute, said the group confirmed with OpenAI before testing that the teen features were live. He added that while some accounts were linked inside that activation window, others had been linked much longer and still produced no alerts. The group also says it shared its results with OpenAI before publishing.

Two things are worth knowing about the relationship. The institute is funded partly by charities and industry, including the OpenAI Foundation, and says it keeps editorial control. And this is not its first warning: it rated ChatGPT “High Risk” overall in October 2025, then a month later called it an unacceptable risk for mental health support.

05

What this means for parents of teens using ChatGPT

The practical takeaway is not to treat the teen mode as a safety net on its own. Even if OpenAI is right about activation delays, the report shows alerts can’t be counted on in the first hours or days after you link an account, which is exactly when many families set things up.

Common Sense Media goes further. It wants OpenAI to stop promoting the teen mode and says teens shouldn’t use ChatGPT at all until independent testing shows the protections work. Many families won’t go that far, but a few steps help either way:

  • Link your account early and check the parental settings rather than assuming alerts are on.
  • Talk with your teen about what they use ChatGPT for, and agree that a chatbot is not the place to handle a mental health crisis.
  • Make sure your teen knows a real crisis line and a trusted adult they can contact.
  • For schoolwork, check whether Study Mode is actually being used, since it can be skipped in seconds.
06

How many teens use it

OpenAI released fresh usage figures the same day. It said nearly 1.2 million teens used ChatGPT’s interactive learning visuals for math and science in a single week, and that fewer than two percent of teen users spent more than three hours in a row on the chatbot. Those numbers describe one feature and one habit, not overall safety, and they don’t answer the alert question at the heart of the report.

07

Quick answers

What does unacceptable risk mean?

It is Common Sense Media’s lowest rating. The group uses it when it believes a product shouldn’t be used by the age group in question, here everyone under 18.

Does ChatGPT for Teens alert parents?

It is designed to, after staff review flagged chats. In Common Sense Media’s tests, a dozen newly linked accounts produced no alerts even after an hour of crisis-related messages.

Has OpenAI changed anything since the report?

OpenAI disputed the methodology and said it will keep working with experts, but it has not announced specific fixes in response.

You're reading the quick version.