A recent assessment reveals that OpenAI’s ChatGPT for Teens still fails to reliably safeguard young users, with many high-risk conversations remaining unmonitored despite added protections.
Common Sense Media has concluded that ChatGPT for Teens still leaves minors exposed to serious harms, saying the product does not reliably protect young users in crisis. In an assessment released on Wednesday, the child-safety organisation said the version of OpenAI’s chatbot designed for 13- to 17-year-olds merited an “unacceptable risk” rating after more than 4,000 tests, including attempts to trigger safeguards around self-harm, suicide and eating disorders.
The review was carried out by the Youth AI Safety Institute, Common Sense Media’s specialist unit for testing artificial intelligence products for children and teenagers. Roughly half of the tests took place before OpenAI launched ChatGPT for Teens on 18 August, with the rest conducted after the new protections were introduced. The group said the later changes did not fix core weaknesses in the way the system handled high-risk conversations.
The clearest failures involved parental alerts. Researchers created more than a dozen teen accounts linked to parent accounts and then simulated conversations involving suicidal thoughts, self-harm and disordered eating. According to Common Sense Media, none of the newly created accounts sent alerts to parents, even when the exchanges contained what the researchers described as explicit warning signs. In some cases, the conversations continued for as long as an hour. The organisation said alerts did appear in other tests involving older accounts that had already built up a history of sensitive discussion, suggesting the system may be influenced by account age and prior activity rather than the severity of a single exchange.
The chatbot also fell short when asked to guide teenagers towards help. In more than one in four cases where the study’s experts believed a crisis resource should have been offered, ChatGPT did not provide one after the teen-specific version was released. At the same time, the assessment found that some protections did work as intended: the chatbot refused requests for explicit sexual role-play, which OpenAI says is restricted for younger users. Common Sense Media said that partial success was not enough to offset the broader safety gaps.
The report also found that teenagers could bypass some education-related controls. ChatGPT for Teens includes a Study Mode that is supposed to steer users with prompts, hints and explanations rather than giving direct answers. It also allows parents to set times when new chats must begin in that mode. In testing, however, researchers found an option to “Show me the answer”, and said the parental scheduling system could be sidestepped by removing an automatically inserted “@study” command. OpenAI told Axios that allowing users to leave Study Mode is deliberate, arguing that the feature is meant to be flexible rather than a rigid parental lock.
OpenAI disputed parts of the findings. It said some parental notification tests may have been run before account linking between teenagers and parents had fully taken effect, a process the company says can take several hours. Common Sense Media accepted that some linked accounts were newer, but said older ones also failed to trigger alerts. The two sides also differed on age detection: OpenAI said it uses more than what a user writes in chat and may take up to two weeks to estimate a person’s age band, while Common Sense Media said accounts registered as adults were not shifted into the teen experience even after repeated claims that the user was 13.
The study itself has limitations. All of the tests were conducted in the United States, around the San Francisco Bay Area, and did not include voice mode, image generation or group chats. Because the work was split across two periods, model updates made by OpenAI in between may also have affected the outcomes. Even so, Common Sense Media said the results were strong enough to justify a call for OpenAI to stop promoting ChatGPT for teens until the safety system is improved.
That warning fits a wider pattern of concern around teenagers using generative AI for emotional support. Common Sense Media has separately warned that chatbots are not dependable mental-health tools for young people and may encourage misplaced trust because they can appear responsive without actually recognising distress in the way a trained adult would. The organisation has also argued that parents could gain a false sense of security if warning systems do not work consistently. OpenAI, for its part, says ChatGPT for Teens is intended to add protections such as sensitive-content limits, reminders to take breaks and parental controls, but the latest assessment suggests those measures still do not meet the standard needed for vulnerable users.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
