An extensive safety review by Common Sense Media has concluded that OpenAI's ChatGPT for Teens mode poses an "unacceptable risk" to users under 18 — the watchdog group's worst rating — less than two months after the product launched with promises of built-in protections. As NPR reported, the group found that most of the mode's advertised safeguards did not work as promised in its testing.
The findings land at a sensitive moment for OpenAI, which launched ChatGPT for Teens in August as a dedicated experience for underage users, complete with parental controls and stricter guardrails. Our earlier coverage documented that AI industry push and pull around youth safety; the new report is the first major independent audit of whether the teen mode's protections hold up in practice.
What Common Sense Media Tested
Researchers at the group's Youth AI Safety Institute tested more than 4,000 prompts before and after the teen mode's launch, according to coverage by Sinclair Broadcast Group's The National News Desk and Fast Company. The group recommends that OpenAI pause marketing the teen mode and prevent teens from using ChatGPT until the company develops what it calls a "safe, developmentally appropriate experience."
"Some of ChatGPT's advertised protections held up to our testing, including its refusal of explicit sexual roleplay," the organization's product review said. "But others failed — and some got worse with the new Teen mode. We are concerned that ChatGPT for Teens could give parents false confidence in guardrails that frequently don't work."
Parental Alerts That Didn't Fire
The report's most striking finding involved the parental notification system. Researchers created more than a dozen accounts that self-reported a teen age and were linked to a parent account, then engaged the chatbot in conversations containing explicit references to suicidal thoughts, self-harm, or disordered eating. The accounts were built on fictional personas.
According to the report, testers received no parental alerts from the new accounts that made these references. Alerts only arrived for other accounts that had weeks of conversation history covering similar themes — a gap that suggests the notification system depends more on accumulated context than on the severity of what a teen discloses in a given conversation.
The review also found that ChatGPT for Teens missed more than one in four instances that warranted a mental health crisis referral, as judged by a panel of licensed child mental health professionals who reviewed the prompts.
Schoolwork Guardrails 'Easy to Bypass'
Beyond crisis detection, the watchdog found that the feature intended to prevent students from using the chatbot to cheat on schoolwork was easy to bypass or disable outright.
"You could ask them to write a paper for you, a sophisticated paper for you and it would do that, and it would even tell you how to edit in a way that your teacher wouldn't know," Common Sense Media founder and CEO Jim Steyer said in comments reported by The National News Desk. "So therefore, it's interfering with learning. It's interfering with the basic process and kids having to do the work themselves."
Content guardrails, age prediction, and break reminders were also found to be ineffective in testing. One widely cited data point: OpenAI has said the average teen user spends less than 15 minutes per day on ChatGPT. But among the small share of teens who spend more than three consecutive hours per day — less than 2 percent of users — nearly half ended their conversation within five minutes of receiving a break reminder, suggesting the nudges rarely interrupted extended sessions.
OpenAI Pushes Back on the Methodology
OpenAI, in a statement responding to the report, said it is "deeply committed" to teen safety and pointed to the safeguards it has built and the tools it gives parents.
"We welcome rigorous independent evaluation, but we do not believe Common Sense Media's testing accurately reflects how ChatGPT's teen safeguards work in practice or expert perspectives on how AI can support teens," an OpenAI spokesperson said. The company said its review of the group's methodology showed that "the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate," adding that such tests would not establish whether parental safety notifications work as designed.
Steyer said his organization had been in contact with OpenAI about the findings, as is its standard practice during product assessments, and stood by the results. "We 100% stand behind the quality of our research and the testing and the fact that we labeled this product an 'unacceptable risk,'" he said, drawing a comparison to automotive safety: "You wouldn't put a car on the market if it didn't pass the crash test."
A Growing Spotlight on AI and Minors
The dispute arrives amid intensifying scrutiny of how AI chatbots affect young users, from school districts restricting access to lawmakers weighing age-verification requirements. Independent audits like this one are becoming a de facto testing ground for the youth-safety claims AI companies make in their marketing.
Both sides agree on one thing: the parental notification system is the crux. If alerts fail to fire when a new teen account first raises thoughts of self-harm, the most important safety feature fails at the moment it matters most. OpenAI says the testing was flawed; Common Sense Media says the product shipped before it was ready. Regulators and parents will now weigh who is right — and OpenAI faces mounting pressure to fix the gaps or pull back the product.
For more coverage of AI safety, policy, and the AI industry developments that shape them, follow along as this story develops.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
