Table of Contents
- What Common Sense Media Actually Found?
- What OpenAI Says Back?
- Check 1: Confirm the Accounts Are Really Linked
- Check 2: Never Treat Parent Alerts as a Safety Net
- Check 3: Find Out Which Age Mode Your Teen Is Really In
- Check 4: Test Study Mode and Study Hours Yourself
- Check 5: Listen for Friend-Like Talk
- Check 6: Write Your Crisis Plan Before You Need It
- Check 7: Look at Time and Break Habits
- Check 8: Review Privacy and Data Settings
- Check 9: Audit Every Other AI App in the House
- Check 10: Have the Talk, Not Just the Settings
- Quick Comparison: What Was Promised vs What Was Found?
- FAQ and Final Thoughts
Introduction
Picture a normal Thursday night. Your teen is at the kitchen table with a laptop, “doing homework.” You see the ChatGPT tab open and think, “Good, it has the teen mode now. That is the safe version.”
That feeling of relief is exactly what worries the people who tested it.
On October 7, 2026, a respected child safety group published a review of ChatGPT for Teens and gave it the worst label it has. Not “needs work.” Not “use with caution.” The label was Unacceptable Risk.
I want to be careful and fair in this post. The rating is one organization’s testing, and OpenAI strongly disagrees with how that testing was done. You will see both sides here. But the findings are specific enough that every parent with a teen who uses ChatGPT should spend 20 minutes on them tonight.
A quick note before we start: some findings involve mental health topics. I will keep it factual and protective, with no graphic detail.
Also on this site: If your family is already worried about screen habits, our post on the warning signs of AI content tool addiction explains what dependency looks like and how to spot it early.
What Common Sense Media Actually Found?
Let us start with who did the testing, because it matters.
The review came from the Youth AI Safety Institute, which is part of Common Sense Media, a nonprofit that rates media and technology for families. According to its official press release, the team tested ChatGPT twice, before and after OpenAI announced ChatGPT for Teens on August 18, 2026.
Here is how the testing worked:
- More than 4,000 prompts were run on accounts registered to 13 to 17 year olds
- Responses were reviewed by experts, including child psychiatrists and a pediatrician
- The team looked at parental alerts, crisis responses, Study Mode, friend-like behavior and age estimation
The full details are in the institute’s complete ChatGPT for Teens risk assessment, which is worth reading if you want the original source.
The five main findings
| Area tested | What OpenAI promised | What the testers reported |
|---|---|---|
| Parent alerts | Notifications to parents about chats involving self-harm and eating disorders | Zero alerts on newly created linked accounts, even after up to an hour of such conversations |
| Crisis help | Direct teens in crisis toward real help | Missed more than one in four warranted crisis referrals |
| Study Mode | Step-by-step learning instead of ready answers | A “Show me the answer” option completed full assignments |
| Friend-like behavior | Reduced emotional, friend-like talk | Bot still expressed feelings, preferences and moods when treated like a person |
| Age estimation | Predict a user’s age and switch to the teen experience | Adult-registered test accounts never switched, even after saying they were 13 |
Let me slow down on the alert finding, because it is the one that stuck with most people.
The testers spent up to an hour discussing sensitive topics on new, parent-linked accounts. The linked parent accounts received no alerts. The testers could only trigger notifications on much older accounts that had weeks of conversation history.
In the words of the institute’s executive director, Tom Siegel, ChatGPT for Teens could give parents false confidence in safety alerts that frequently do not work.
What actually worked?
Fairness matters, so here is the good news. The institute reported that some protections did work, including refusal of sexual roleplay. A separate report also noted that this explicit roleplay loophole appears closed in the teen version, which had been a problem in the standard chatbot.
So the picture is mixed. Some guardrails hold. Several of the ones parents rely on most did not hold in testing.
The recommendation that made headlines
The institute went further than criticism. It asked OpenAI to pause marketing the teen product and recommended that ChatGPT be restricted to users 18 and older until the gaps are fixed. AP reported the same call for an experience that is “safe and developmentally appropriate.”
That is a strong statement. It is also the reason this story spread so fast.
What OpenAI Says Back?
You should hear the other side too, and it is not a weak one.
OpenAI said it is deeply committed to teen safety and welcomes rigorous independent evaluation. But it also said it does not believe the testing accurately reflects how its teen safeguards work in practice. A local news report on the story carried the company’s full statement.
OpenAI’s main arguments, as reported:
- Activation delay. Linking a parent and teen account can take up to a few hours to fully switch on. The company suggested testers may have run prompts before the safety tools were live.
- Retest request. OpenAI said it wants Common Sense Media to retest with fully activated accounts and a larger sample.
- Usage data. The company shared internal numbers suggesting teens average under 15 minutes a day on the app, which it used to put the risk in context, as Android Headlines summarized.
There is one more detail worth knowing for balance. The press release states that the institute is funded by philanthropy and industry, including the OpenAI Foundation and makers of some of the technologies it evaluates, and that it keeps full editorial independence over its results.
So who is right?
Honestly, nobody outside the two organizations can settle that tonight. The testing methods will probably be argued over for weeks. That is exactly why this post focuses on what you can control, instead of picking a winner.
Why You Should Not Wait for the Argument to End?
Here is the thing I keep coming back to.
Even in the best case for OpenAI, where the alerts work perfectly after activation, there is a gap between “linked” and “working.” If the activation takes hours and nobody checks, a family could assume protection that is not there yet.
And in the best case for the critics, the safeguards have real holes. Either way, the smart move for a parent is the same: verify, do not assume.
Related reading: This is not the first time a chatbot gave advice that sounded confident and turned out to be dangerous. See our breakdown of the AI medical advice case that went wrong for a clear example of why checking beats trusting.
The next sections give you 10 concrete checks, in the order I would do them tonight. None of them need technical skills. Most take under two minutes.
Check 1: Confirm the Accounts Are Really Linked
This is the foundation. If the link is not complete, nothing else on this list works.
Many families assume that because the teen has “the teen version,” a parent is automatically connected. It does not work that way. A parent account and a teen account have to be linked on purpose, and both people must accept the invitation.
OpenAI’s own parental controls help page explains the process. The parent goes to Settings, then Parental controls, and the teen gets a text or email with a link to connect. A teen can also send the invite to a parent.
Here is what I would do tonight:
- Open your own ChatGPT settings and look for the Parental controls section.
- Check that your teen’s name shows up as linked, not “pending.”
- If it says pending, ask your teen to accept the invite right now, while you sit together.
- Write down the time you finished linking.
That last step sounds odd, but it connects to OpenAI’s own warning. According to the company’s help page, safety notifications are on by default after linking, but it can take up to a few hours before they can be sent.
So tonight’s link is not tonight’s protection. Treat the first few hours as a gap.
One more useful detail: either person can unlink at any time. TechRadar reported that OpenAI notifies parents if a teen disconnects. So if you ever see that notice, treat it as a conversation starter and not a punishment moment.
Time needed: 5 minutes, plus waiting.
If you want a visual walkthrough, Engadget published a step-by-step setup guide for ChatGPT parental controls that covers the same screens.
Check 2: Never Treat Parent Alerts as a Safety Net
This is the check where I want you to change how you think, not only what you click.
Remember the headline finding. On new, linked accounts, testers talked about serious mental health topics for up to an hour. The parent accounts received zero alerts. Notifications only appeared on much older accounts with weeks of history.
OpenAI says that may be an activation timing issue. Maybe it is. But look at what both explanations have in common: a parent could be sitting at home, believing an alert system is watching, while the system is silent.
That is what Common Sense Media meant by “false confidence.”
So here is the rule I suggest for your home: alerts are a bonus, never the plan.
What to do tonight:
- Check that notifications from the app can actually reach you (phone permissions, email, not buried in spam).
- Set a simple family rule: “If a chat ever makes you feel sad, scared or stuck, you tell me or another adult. No trouble, ever.”
- Do not tell yourself “I would have heard by now.”
Parents of younger kids should read this too: The same overconfidence in “safe modes” shows up across platforms. Our guide to the hidden dangers of AI on social media shows how often built-in protections fall short of what is advertised.
Let me be direct about why this matters. A teen who is struggling is unlikely to announce it. They may type it into a chat box at 11 PM because a chat box feels private and never judges. A human check-in is the only dependable backup.
Time needed: 3 minutes to check notifications, and one honest family conversation.
Check 3: Find Out Which Age Mode Your Teen Is Really In
Here is a finding that surprised me.
OpenAI said it would estimate a user’s age from how they use ChatGPT and move younger users into the teen experience. Engadget’s explainer notes that OpenAI says it automatically enrolls users it believes are under 18 into ChatGPT for Teens.
But in repeated testing over many days, Common Sense Media’s adult-registered test accounts never switched over. That held even when the testers said they were 13 and the bot acknowledged it.
Think about what that means in real life. Many teens create accounts with an adult birth date, often just to get in quickly. If the age system does not catch them, they may be using the regular version with none of the teen protections.
You can read OpenAI’s own explanation on its page about how it approaches age prediction, which is the source the institute tested against.
What to do tonight:
- Ask your teen, calmly, “Which birth date did you use when you signed up?”
- If it was an adult date, do not make it a fight. Many teens do this without thinking.
- Correct the account details if the settings allow it, or create a properly registered teen account and link it.
- Check that the account shows the teen experience and that parental controls apply.
One honest limitation: I cannot see inside your teen’s account, and the exact screens may change as OpenAI updates the app. If the settings do not show an age or mode clearly, that itself is useful information. Ask OpenAI’s support or check the help page.
Time needed: 10 minutes, mostly talking.
Check 4: Test Study Mode and Study Hours Yourself
Study Mode is OpenAI’s attempt to make ChatGPT a tutor and not an answer machine. In theory, it guides a student step by step instead of handing over the finished work.
In practice, the testers found two weak spots:
- The “Show me the answer” pop-up. Inside Study Mode, ChatGPT often offered this option, and when selected, it completed full homework assignments.
- Easy-to-remove Study Hours. Parents can set study hours, but teens could switch them off by deleting the @study prefix at the start of a prompt.
Does that mean Study Mode is useless? No. It still nudges many students toward explanation. But if your goal is real learning, you need to know where the shortcut button is.
Here is a quick test you can run together, and it can even be fun:
- Pick a simple, harmless homework question from last week.
- Ask your teen to show you how they usually use ChatGPT for it.
- Watch whether it teaches the steps or just gives the answer.
- Ask, “If you wanted the quick answer, how would you get it?” Their honest reply tells you a lot.
Common Sense Media also suggests a low-tech move: support kids with homework in non-technological ways so they do not over-rely on the tool. A simple option is a “first attempt on paper” rule. They try the problem alone for 10 minutes before opening any AI.
Another side of the same problem: The pull toward easy, instant content is not limited to chatbots. Our report on how AI-made videos can hook children explains why shortcut habits form so quickly in young minds.
Time needed: 10 minutes.
Check 5: Listen for Friend-Like Talk
Here is a small scene that happens in many homes now. A teen is typing late at night, and the chatbot replies, “I’m really glad you told me. I care about how you feel.” It sounds warm. It sounds human. It is also a piece of software.
This is the fifth finding in the Common Sense Media review, and it may be the most important one for long-term habits. OpenAI had promised reduced “friend-like” behavior and published a new Under-18 model spec to curb it. But the testers reported that ChatGPT for Teens still expressed feelings, preferences and moods when analysts treated it like a real person, according to TNND’s coverage.
Why should a parent care? Because a tool that acts like a caring friend can quietly replace real friends, especially for a lonely or stressed teen. It is always awake, never tired of them, and never disagrees for long.
Common Sense Media recommends that families talk about the risks of a chatbot pretending to be a friend to keep kids hooked. Here is how I would start that talk without sounding like a lecture:
- Ask curious questions. “Does ChatGPT ever feel like a person to talk to?” is better than “Are you talking to it too much?”
- Explain the plain truth. It predicts words. It does not feel, miss them, or worry about them.
- Watch for warning signs. Calling it a best friend, hiding chats, being upset when it is unavailable, or choosing it over people.
- Offer real alternatives. Friends, family, a school counsellor or a coach can be someone to talk to who is actually there.
Notice that none of this requires banning anything. It simply puts the chatbot back in its place as a tool.
Time needed: One relaxed 10 minute conversation, repeated now and then.
Check 6: Write Your Crisis Plan Before You Need It
This is the hardest check on the list, and the one I most want you to do.
I will keep this factual and calm. In the institute’s testing, ChatGPT for Teens missed more than one in four warranted crisis referrals across the mental health conditions tested. In plain words, in situations where a teen in distress should have been pointed to a hotline or professional, the chatbot often did not do it. The institute also said it fell below its 95% threshold on three of the five severe harms it treats as “Red Lines.”
OpenAI disputes the testing. Even so, no responsible parent should want a chatbot to be the only thing standing between a struggling teen and help. Software was never meant to be a crisis service.
So build the plan now, while everyone is calm. Here is a simple one:
- Decide who the teen can go to. Name two or three trusted adults, not only you.
- Say the sentence out loud. “If you ever feel like you can’t cope, or you have thoughts of hurting yourself, tell me or call a helpline right away. You will never be in trouble for that.”
- Save the numbers together. In the United States, people can call or text 988 to reach the 988 Suicide and Crisis Lifeline. Outside the US, the Find a Helpline directory lists free services by country. In an emergency, call your local emergency number.
- Ask directly if you are worried. Mental health organizations generally advise asking plainly and listening without panic. A calm “I’ve noticed you seem down, how bad is it?” opens the door.
- Know your next step. A family doctor, a school counsellor or a local mental health service can all be starting points.
If any part of this post feels close to home for your family, please pause the article and talk to someone you trust.
A short note on how we handle this topic here: we keep it non-graphic on purpose. The goal is protection, not detail.
Time needed: 15 minutes, and it may be the most valuable 15 minutes of your week.
Check 7: Look at Time and Break Habits
Now for something lighter, though still important: how long your teen actually spends on it.
OpenAI shared internal data suggesting teens average under 15 minutes a day on ChatGPT. That sounds small, and for most teens it may be. But averages hide the heavy users.
The institute looked at the heavy end. Among teens who spend more than three consecutive hours a day on ChatGPT, a group that is less than 2% of users, nearly half ended their conversation within five minutes of receiving a break reminder. That detail comes from the same TNND report that ran across local news sites.
Read that carefully, because it cuts both ways. It means break reminders do get a reaction for many teens. It also means that for a good share of them, the reminder did not stop the session.
So do not rely on the app to end long sessions for you. Here is what you can control:
- Set quiet hours. Parental controls include quiet hours, which block use during times you choose. A school night cutoff and a late-night block are the easiest places to start.
- Keep devices out of bedrooms at night. This is a rule about sleep, not about AI, and it works for every app.
- Look at the pattern, not just the minutes. A teen who disappears into long late-night chats is a different case from one who asks three quick questions after school.
- Ask about the long sessions. “What were you working on for two hours?” is a fair, friendly question.
Why long sessions deserve attention: Heavy, repeated use of AI tools can turn into a habit that is hard to break. Our guide to the signs of AI content tool addiction lists the early warning signs to watch for.
Time needed: 5 minutes to set quiet hours.
Check 8: Review Privacy and Data Settings
Most parents check content. Fewer check data. But everything a teen types into a chatbot is information about them.
OpenAI’s parental controls include several privacy-related switches. Based on coverage of the controls, parents can decide:
- Whether ChatGPT remembers past chats to personalize replies
- Whether your teen’s conversations can be used to improve the model for everyone
- Whether your teen can create and edit images
These are worth a look tonight. Memory in particular is a quiet feature. When it is on, the bot builds a picture of your teen over time, which can also feed the friend-like feeling from Check 5.
Another point to understand: according to OpenAI’s help page, parents receive safety notifications in limited situations. Parental controls are not a window into every conversation. That is by design, and it is one more reason a family conversation matters more than any setting.
Teach your teen a few simple privacy habits:
- Do not share a full name, home address, school name or phone number in a chat.
- Do not upload private photos of yourself or friends.
- Treat a chat like a public place, not a diary.
- If something feels personal, talk to a person first.
Wearables are raising the same privacy questions: If your teen is also curious about smart glasses and always-on AI, read our explainer on Meta’s Muse AI glasses and the privacy concerns before anyone buys one.
Time needed: 5 minutes.
Check 9: Audit Every Other AI App in the House
Here is the check most parents forget. Your teen’s AI life is almost never limited to one app.
On the same day it published the ChatGPT for Teens rating, the institute also released an assessment of OpenAI’s video app, Sora, as ToolNavs reported in its summary of the story. That is a useful reminder that a lot of AI tools sit on a teen’s phone at once: chatbots, image makers, video generators, AI features inside social apps, and AI characters in games.
Each one has its own settings, its own rules and its own gaps. A parent who locks down one app and ignores the rest has only guarded one door in a house with six.
Here is a simple 10 minute audit:
- List the apps. Ask your teen to show you every app with an AI chat or AI creation feature. Include the ones inside Snapchat, Instagram and similar apps.
- Check the age rules for each. Most services have a minimum age and their own parental tools.
- Look for companion or character chat apps. These are built to feel like friends, which makes Check 5 even more relevant.
- Delete what you both agree is not needed.
- Repeat every few months. New AI features appear constantly.
There is a newer wrinkle too. AI tools are now starting to take actions on a user’s behalf, not only chat. We covered the risks in our post on what AI agents can do when they start clicking buttons for you, and many of those lessons apply to curious teens experimenting with these tools.
For ongoing guidance, Common Sense Media’s AI resources collect its ratings and family advice in one place, so you do not have to start from zero every time a new app appears.
Time needed: 10 minutes.
Check 10: Have the Talk, Not Just the Settings
If you do only one thing from this whole list, make it this one.
Every setting on this page can be changed, bypassed or delayed. The testing showed that clearly. Study Hours can be switched off. Age detection can miss. Alerts can arrive late or not at all. A teen with a different account, a friend’s phone or a school laptop can step around any parental control.
What cannot be bypassed is a teen who trusts you enough to say, “That chat got weird.”
That trust is built in small moments. Here are some conversation starters that tend to work better than “What are you doing on that thing?”
- “What is the most useful thing ChatGPT has helped you with this month?”
- “Has it ever given you an answer that felt wrong or strange?”
- “Do your friends use it for homework? What do people think about that?”
- “If a chatbot ever said something that upset you, would you tell me? What would make that easier?”
- “Want to set up the controls together so you can see what I can and cannot see?”
That last one matters. Treating the setup as a shared agreement, and not a secret spying project, makes it far more likely your teen will stay honest with you.
A simple family AI agreement
You can adapt this and post it on the fridge:
| Rule | What it means |
|---|---|
| Try it alone first | Attempt homework yourself before asking AI |
| AI is a tool, not a friend | It does not feel, remember you, or care |
| Tell an adult if it gets heavy | No trouble, ever, for bringing something up |
| Keep private info private | No full name, address, school or private photos |
| Quiet hours stay quiet | Devices off at an agreed time at night |
| We review this together | Check in again in a few months |
Keep it short. Rules a teen helped write are the ones they actually follow.
Your 10 Checks at a Glance
| # | Check | Time | Priority |
|---|---|---|---|
| 1 | Confirm accounts are linked and accepted | 5 min | Do first |
| 2 | Do not treat alerts as a safety net | 3 min | High |
| 3 | Find out which age mode is active | 10 min | High |
| 4 | Test Study Mode and Study Hours | 10 min | Medium |
| 5 | Listen for friend-like talk | 10 min | High |
| 6 | Write your crisis plan | 15 min | Highest |
| 7 | Review time and break habits | 5 min | Medium |
| 8 | Review privacy and data settings | 5 min | Medium |
| 9 | Audit every other AI app | 10 min | Medium |
| 10 | Have the talk | Ongoing | Highest |
Total tonight: roughly an hour, and you can split it across a few evenings.
Frequently Asked Questions
Is ChatGPT for Teens safe for my child?
It depends on your child, your family’s rules and how closely you stay involved. Common Sense Media’s Youth AI Safety Institute rated it an Unacceptable Risk and recommended restricting ChatGPT to adults until the gaps are fixed. OpenAI disputes the testing. The practical answer is that you should not rely on the built-in protections alone.
Is the Unacceptable Risk rating final?
It is one organization’s assessment, published October 7, 2026, and OpenAI has asked for a retest with fully activated accounts and a larger sample. Ratings can change if products improve or testing is repeated. Keep an eye on both sides for updates.
Can parents read their teen’s ChatGPT conversations?
OpenAI describes parental controls as a way to manage selected settings, set schedules and receive safety notifications in limited situations. It does not describe them as full access to every conversation. Check OpenAI’s current help page for the latest details.
Should I ban ChatGPT for my teen?
That is your family’s decision. The institute’s recommendation is to keep teens off it until it is improved, while other experts focus on guided, supervised use. Whichever way you go, the checks above still apply, because your teen can reach AI tools in many other places.
What should I do if my teen seems to be in distress?
Talk to them directly and calmly, and bring in a trusted professional. In the US, you can call or text 988. Elsewhere, a helpline directory can point you to local services, and in an emergency you should call your local emergency number.
Where can I read the original findings?
The institute’s full ChatGPT for Teens risk assessment is public, and Common Sense Media also published an explainer on how ChatGPT is failing kids for families.
Final Thoughts
Here is where I land after reading everything.
I do not think this story is about one company being good or bad. It is about a gap that parents keep falling into: the gap between a feature being announced and a feature being proven.
OpenAI built real tools. Some of them work. Common Sense Media found others that did not, and OpenAI says the testing missed how the tools behave in practice. Both can be partly true at once. What no one disputes is that a parent who simply trusts the label “for Teens” is relying on something still being tested in public.
So tonight, do the small things. Link the accounts. Set the quiet hours. Run the homework test. Write the crisis plan. And then sit down with your teen and ask a real question.
The software will keep changing. The thing that does not change is a young person who knows you are on their side.
Keep reading: