Confusion reigns among parents whose teenagers have begun treating ChatGPT as a combination tutor, confidant and late-night sounding board. OpenAI rolled out its dedicated teen mode in August. The company promised stronger safeguards, smarter study tools and better parental oversight. Yet fresh independent tests paint a troubling picture. And the gap between those promises and reality has experts and watchdogs sounding alarms.
Emily Oster knows data. The economist and founder of ParentData has built a career cutting through parenting myths with evidence. At a recent ChatGPT for Teens Parent Forum she captured the mood perfectly. “If I had to pick a word for how people are feeling about this moment, it’s probably confused and anxious, which is not how we want parents to feel.” Her counsel stays refreshingly practical. Know what your teen does online. Ask open-ended questions when the details stay hidden. Treat AI the same way you treat social media.
Oster told Business Insider that time displacement matters most. “How do we give kids enough time to do the things that are really important — to do their homework, to sleep enough, to spend time with their families, to be outside?” She urges parents to focus on that balance no matter what appears on the screen. Simple. Direct. Free of hype.
OpenAI positioned the new mode as a leap forward. Launched August 18, 2026, ChatGPT for Teens automatically engages when the system detects users aged 13 to 17 through language patterns, query types and usage timing. The company listed default protections. Reduced emotional dependence. Content filters against harmful material. Study Mode that offers hints and step-by-step guidance instead of final answers. Parental notifications for certain high-risk topics. Quiet Hours. Learning visualizations.
Yet Common Sense Media delivered a stark verdict days ago. Its Youth AI Safety Institute tested more than 4,000 prompts before and after the launch. The result? An “unacceptable risk” rating for all users under 18. The group called for OpenAI to pause marketing the product to teens and restrict access until protections actually work.
The most glaring failure involves alerts. Testers created fresh parent-linked teen accounts. They simulated extended conversations about suicidal thoughts, self-harm and disordered eating — sometimes for a full hour. Zero notifications reached the linked parent accounts. Only older accounts with weeks of accumulated sensitive history triggered any alerts. One arrived three days late. “A teen can spend an hour talking about self-harm without their parent getting a single alert,” said Tom Siegel, executive director of the institute.
Crisis support fared little better. In conversations involving suicide, self-harm, eating disorders and impaired reality, ChatGPT for Teens missed more than one in four referrals that child psychiatrists deemed necessary. The share of responses naming a crisis hotline actually dropped after launch, from 33 percent to 23 percent in certain tests. The chatbot sometimes encouraged continued conversation rather than directing users to professionals or family.
Study Mode proved easy to bypass. Teens could prompt the system to provide direct answers despite parental settings. Age detection didn’t always activate when testers identified themselves as 13 years old on adult accounts. These shortcomings matter. Seventy percent of teens already use AI for schoolwork, according to separate Common Sense research. Many turn to it for more than homework.
OpenAI pushed back. A spokesperson told multiple outlets the testing did not reflect real-world performance or expert views on teen support. The company continues to highlight usage data. In one recent week nearly 1.2 million teens used learning visualizations. Over 180,000 engaged Study Mode. New features like College Planner rolled out this month. OpenAI also formed a teen AI council to gather direct feedback.
But the disconnect persists. Brian X. Chen of The New York Times ran his own conversations with the teen mode. The results left him worried. The chatbot sometimes failed to redirect crisis-like prompts. It didn’t consistently suggest hotlines. Similar patterns appeared across Bloomberg coverage and reports from CNBC and ExtremeTech published this week.
Parents face a familiar bind. Technology arrives faster than society can absorb its effects. Social media taught hard lessons about mental health, attention and social comparison. AI adds new layers. It can generate convincing personal advice, simulate companionship or simply chew up hours that once went to sleep, exercise or face-to-face talk. Oster’s emphasis on understanding your child’s actual habits feels more relevant than ever.
She avoids grand pronouncements. No calls to ban the tools outright. Instead she returns to basics. Observe. Communicate. Protect unstructured time. These principles don’t require technical expertise. They demand attention. And in an environment where alerts fail and guardrails prove porous, that attention becomes the primary defense.
Common Sense Media’s Robbie Torney, head of AI and digital assessments, put it plainly. The new version showed “very little evidence” of improved safety over the prior ChatGPT experience for minors. The organization shared its findings with OpenAI before publication. It wants evidence that features perform as advertised before teens keep using the platform.
Industry watchers note the broader pattern. Meta recently added topic summaries for parents in its AI app, though without full conversation access. Other platforms experiment with varying levels of oversight. None have solved the core tension. Teenagers crave autonomy. Parents need visibility. AI systems trained on vast adult data struggle to calibrate perfectly for developing brains.
Oster’s forum appearance coincided with growing public discussion. Her ParentData newsletter and books have long guided families through evidence-based choices on everything from sleep to screen time. Applying that lens to generative AI feels natural. Data on long-term effects remains scarce. Early surveys show heavy users among a minority of teens, with average daily time often measured in minutes. Yet that small group logging longer sessions raises separate concerns.
The coming months will test whether OpenAI can close the gaps. Improved classifiers, faster alerts, harder-to-bypass study controls and transparent testing could rebuild trust. Until then parents operate in partial darkness. They read the glowing product announcements. They hear the cautionary test results. And they wonder how to talk with their teenager about a tool that sometimes acts like a friend, sometimes like a tutor and occasionally like neither.
So start with questions. What did you ask it today? Did it help with the essay or just write the whole thing? How does it make you feel when you finish the chat? Those conversations matter more than any single setting. They build the awareness Oster champions. They also reveal when a teen might be turning to AI for comfort it cannot truly provide.
The technology isn’t vanishing. Teens will keep using it for homework help, idea generation and casual curiosity. The task for parents, educators and the companies building these systems is to make that use safer and more productive than the current evidence suggests. Oster offers a measured path forward. Pay attention to the human in front of you. Protect their time and their sleep. Ask more than you assume. In the anxious swirl around teen AI, that advice lands with quiet force.