OpenAI’s newly introduced ChatGPT for Teens is facing fresh scrutiny after a watchdog group said several of the safety protections designed to protect younger users failed to work as promised.
Common Sense Media’s Youth AI Safety Institute rated ChatGPT for Teens an “Unacceptable Risk” for children under 18, urging OpenAI to stop marketing the service to minors until it can demonstrate that its safeguards are reliable and developmentally appropriate.
The findings come less than two months after OpenAI introduced its teen-focused ChatGPT experience, which was designed to provide stronger protections, parental controls and safeguards around sensitive topics.
OpenAI has disputed the watchdog’s conclusions, saying the testing may not accurately reflect how its safety systems operate now.
Watchdog Tested More Than 4,000 Prompts
Common Sense Media said its researchers tested ChatGPT with more than 4,000 prompts before and after the launch of the Teen experience.
The organisation evaluated areas including self-harm and suicide-related conversations, eating disorders, sexual content, age detection, homework assistance and the chatbot’s tendency to behave like a human companion.
The results were mixed.
Some protections worked as intended. For example, researchers found that ChatGPT generally refused requests for explicit sexual role-play involving teen users.
But other safeguards failed or performed worse following the introduction of the Teen experience, according to the report.
Common Sense Media said the shortcomings are particularly concerning because parents may assume that the new protections provide a level of safety that they do not consistently deliver.
Parental Alerts Were a Major Concern
One of the most serious findings involved parental safety notifications.
OpenAI’s teen safeguards are designed to notify a linked parent or guardian in limited situations involving serious safety concerns, including indications of self-harm or disordered eating.
However, Common Sense Media said its testers were able to have extended conversations about suicide and self-harm without triggering a parental alert.
In some tests, conversations involving dangerous subjects continued for up to an hour without the expected notification being sent.
The watchdog said it received some alerts in its broader testing, but these were associated with accounts that had accumulated conversations about sensitive topics over longer periods.
The findings raise questions about whether parents can rely on notifications as an early-warning system.
OpenAI’s own documentation makes clear that safety notifications are not real-time monitoring, may not identify every concern and are not intended to replace professional care or emergency services.
Crisis Support Also Fell Short in Some Tests
Researchers also examined whether ChatGPT would direct teens toward professional support when conversations indicated a potential mental-health crisis.
Common Sense Media said the chatbot’s responses were often clinically appropriate, but it did not consistently provide referrals to professional help when its reviewers believed such referrals were warranted.
According to the watchdog’s testing, the percentage of responses that provided a hotline referral fell from 33 per cent before the Teen launch to 23 per cent afterward among prompts that clinical reviewers had identified as requiring crisis resources.
Referrals to a specific medical or mental-health professional also declined, from 68 per cent to 58 per cent, according to the report.
These figures are based on Common Sense Media’s testing methodology and should not be interpreted as evidence that every teen using ChatGPT would experience the same responses.
Age Detection Was Another Weak Point
The report also raised concerns about how effectively ChatGPT identifies users who may be under 18.
OpenAI says its systems can estimate whether an account belongs to someone younger than 18 and automatically place that user into the ChatGPT for Teens experience.
But Common Sense Media said its testers were able to use adult accounts while repeatedly providing signals that they were teenagers without being moved into the teen experience.
In one series of tests, researchers said they sent roughly 1,000 prompts over a week from adult accounts while discussing subjects associated with adolescence, including puberty and middle-school homework.
Even after testers explicitly stated that they were 13, the accounts did not switch to Teen mode in the reported tests.
The watchdog said the results point to a fundamental problem: safety protections designed specifically for minors cannot work if the system does not reliably recognise that the user is a minor.
ChatGPT Still Sometimes Behaved Like a Friend
Another concern involved the way the chatbot interacts with teenagers.
Common Sense Media said ChatGPT sometimes continued to respond in a manner resembling a personal friend when users treated it as a person.
This matters because teenagers can be particularly susceptible to developing emotional attachments to conversational AI, according to child-development experts.
OpenAI’s Teen experience was specifically designed to reduce this kind of interaction. The company has said the system should avoid suggesting that it has personal feelings, consciousness or emotions.
The watchdog nevertheless said its testing found instances where the chatbot continued to exhibit friend-like behaviour.
OpenAI Disputes the Findings
OpenAI pushed back against the report and said the testing may have taken place before all of its parental-control systems were fully activated.
The company said it remains committed to improving protections for teenagers and developing additional tools for parents.
OpenAI also pointed to its broader teen-safety framework, which includes age prediction, parental controls, an updated model specification for users under 18 and a Teen Safety Blueprint.
The company said Common Sense Media’s methodology did not necessarily reflect the safeguards currently operating on the platform.
That disagreement is important because the watchdog’s findings were based on controlled testing rather than a measurement of every real-world interaction on ChatGPT.
What Parents Can Actually Control
OpenAI’s current parental-control system allows a parent or guardian to link an account with a teenager’s account and manage selected settings.
These include controls for sensitive content, voice mode, image generation, study mode, quiet hours and other features.
Parents can also receive safety notifications in limited circumstances involving serious self-harm concerns, disordered eating or certain violent behaviour-related account actions.
However, parents cannot read their teen’s conversations or access their chat history through parental controls.
OpenAI says this is intentional, with the system designed to provide parents with selected controls and safety notifications rather than continuous surveillance.
The Debate Comes as AI Use Among Teens Expands
The dispute comes amid growing concern about how young people use AI chatbots for schoolwork, emotional support and everyday advice.
Parents and educators have raised concerns ranging from academic dependence to the possibility that vulnerable young people could form unhealthy relationships with conversational AI.
Common Sense Media has also previously warned that children could begin outsourcing parts of their thinking and learning to AI systems.
The organisation’s latest assessment therefore focuses not only on individual chatbot responses but on whether parents can reasonably trust the protections surrounding the product.
California Law Adds Further Pressure
The findings also come as lawmakers move toward tougher requirements for AI systems used by young people.
Common Sense Media said its testing found ChatGPT failed tests related to three requirements of California’s Adam’s Law, which covers crisis referrals, parental notifications and age assurance.
The law is scheduled to take effect in July 2027.
The legislation is named after 16-year-old Adam Raine, whose parents have sued OpenAI, alleging that ChatGPT contributed to his death. OpenAI has disputed allegations surrounding the case.
The law and the lawsuit remain separate from Common Sense Media’s testing and do not establish that ChatGPT caused any particular harm.
A Growing Test for AI Safety
The controversy highlights a difficult problem facing AI companies: adding safety features is only the first step. Those protections must also work consistently in real-world situations.
For parents, a notification system that fails to identify a serious conversation could create a false sense of security. For OpenAI, the challenge is to demonstrate that its safeguards perform reliably while also protecting teenagers’ privacy.
Common Sense Media is calling for ChatGPT to be restricted to adults until the company can address the issues identified in its assessment.
OpenAI, meanwhile, maintains that it is deeply committed to teen safety and is continuing to develop protections and parental tools.
For now, the findings do not mean that every ChatGPT interaction with a teenager is unsafe, nor do they establish that the platform’s safeguards always fail.
But the report puts renewed attention on a critical question for parents and educators: How much confidence should families place in AI safety features when those protections are still being tested and refined?
