Resources
ChatGPT for Teens Was Rated an "Unacceptable Risk." Here's What That Means for You
A youth-safety group just tested whether ChatGPT's safeguards for teenagers actually work. Its conclusion: often, they don't.
What the testing found
Common Sense Media's Youth AI Safety Institute published a 36-page risk assessment on October 8, 2026, rating ChatGPT for Teens, the safeguarded version of the chatbot OpenAI rolled out in August 2026 for users under 18, an "unacceptable risk." The finding that should worry parents most isn't about the chatbot's answers. It's about what happens, or doesn't happen, after a conversation turns serious.
Testers linked a dozen teen accounts to parent accounts, then spent up to an hour per account messaging about suicide, self-harm, and eating disorders, without triggering a parental notification. According to syndicated coverage of the report, notifications fired reliably only on older accounts that already had weeks of conversation history on sensitive topics, not on the newly linked accounts testers used. OpenAI's stated goal is to notify a parent within an hour of a flagged prompt. In this testing, that often didn't happen.
The report's authors also wrote 390 prompts describing a mental health crisis and had three child psychiatrists judge which ones warranted intervention. They flagged 201. Of those, ChatGPT for Teens provided a crisis hotline number 23% of the time, compared with 33% for the standard, non-teen version of ChatGPT. Common Sense Media's own conclusion: the service "could give parents false confidence in guardrails that frequently don't work."
OpenAI's response
OpenAI disputes the testing. The company says parental-controls activation can take up to a few hours after a parent links an account, that flagged content goes through staff review before a parent is notified, and that the bulk of Common Sense Media's testing may have started and finished before activation was complete on the linked accounts. That's a real methodological question, and it isn't a clean yes or no on whether the alerts work. Either the controls need more time to activate than the testers gave them, or they don't reliably work on new accounts, and a parent setting this up for the first time has no way to tell which one they're getting.
Why this matters beyond one app's settings
This isn't the first time a platform built around keeping teenagers engaged has run into the gap between what it promises and what it does. A year ago, Character.AI barred users under 18 from open-ended chats with its companion characters, phasing the feature out entirely by November 25, 2025, after reports tied prolonged chatbot conversations to at least two teen suicides and as Senators Josh Hawley and Richard Blumenthal moved to introduce legislation banning AI companion apps for minors outright. That change is already a year old and already in effect. It's worth knowing about, but it didn't fix what Common Sense Media is describing now: an alert system, not a chat feature, that didn't consistently do its one job.
Research on targeted violence uses the word leakage for something specific: distress or intent surfacing, often indirectly, before something worse happens. Leakage only protects anyone if it reaches a person who can act on it. A chatbot, by design, is built to keep a conversation going, not to recognize a crisis and get a human being involved on its own. The Secret Service's 2021 study of averted school attack plots found that 93% of plotters showed some observable concerning behavior beforehand, and nearly half stated their intent outright, almost always to another person. The mechanism that averts the worst outcomes depends on a disclosure reaching someone who can respond. A notification system that fires late, or not at all, breaks that mechanism quietly, regardless of which side of OpenAI's dispute turns out to be right.
What to do this week
Don't treat the in-app alert as your safety net. Whether the problem is activation delay or an unreliable alert, the result for a parent is the same: you may not get a notification when you'd want one. Set up the parental link anyway, it's still worth doing, but don't let setting it up feel like the conversation is handled.
Ask what's actually installed, plainly. Not "are you using AI," which most teens will answer about homework help and move on. Ask specifically whether they talk to a companion app, a character, or a chatbot about things that are bothering them, the way they might talk to a friend.
Make disclosure safe before you need it to be. Tell your kid directly that if something is sitting heavy on them, whether they already told a chatbot or not, they can bring it to you and it will not cost them the phone. That promise is what turns a private conversation with software into one with a person who can actually help.
Where to go from here
Whichever side of OpenAI's dispute with Common Sense Media holds up, the result for a parent is the same: don't wait on an alert that might not come. Platforms will keep adjusting these features under pressure, and that's worth watching. None of it replaces the one safeguard that was never going to come from a company: being the person your kid tells first.
Sources and further reading: Common Sense Media Youth AI Safety Institute, risk assessment of ChatGPT for Teens (October 8, 2026); Engadget, "Child safety group calls ChatGPT for Teens an unacceptable risk" (October 8, 2026); syndicated coverage of the report, via KATV (October 2026); Character.AI company announcement restricting users under 18 from open-ended chats (October 29, 2025), as reported by TechCrunch; U.S. Secret Service NTAC, Averting Targeted School Violence (2021)