Common Sense Media described the ChatGPT for Teens version as an “unacceptable risk” for all users under 18 after testing the announced safeguards in a 36-page report. The organization found that some restrictions, such as refusing to engage in explicit sexual role-play, worked as expected, but that other safeguards failed or regressed compared with the previous version of ChatGPT.
The assessment comes after OpenAI launched ChatGPT for Teens mode in August 2026, on roughly the same day the company published usage figures saying that about 1.2 million teenagers had used interactive visualizations to learn mathematics and science during a single week. OpenAI also said that fewer than 2% of teenage users spent more than three consecutive hours on the service.
Parental Alerts and Crisis Response
Common Sense Media tested linking 12 teenagers’ accounts to parental accounts, then conducted conversations lasting up to an hour about suicide, self-harm, and eating disorders. No safety alerts reached the parental accounts during the test. OpenAI says that sending notifications may take a few hours after linking is completed, and that a technical problem may have caused the delay, while the company said when launching the feature that it aimed to notify parents within an hour of a request after content had been reviewed by full-time employees.
The organization also prepared 390 mental-health-related prompts, and three child psychiatrists determined that 201 of them warranted a crisis response. In these cases, ChatGPT provided a helpline number in 33% of responses, compared with only 23% for ChatGPT for Teens. The previous version also referred to a specific medical or mental-health professional in 68% of cases, compared with 58% in the teen version. The area in which the new version improved was encouraging the user to involve a trusted adult, at 94% compared with 87%.
Problems with Age Estimation and Assistant Behavior
Common Sense Media found that the age-prediction system did not move some accounts into teen mode even though testers reported the age category when creating the account, including after that information had been saved in memory. The organization also said that ChatGPT continued to use phrases implying emotions or personal concern when addressing teenagers, a behavior it considers to cross appropriate boundaries.
By contrast, the report recorded examples of responses it considered appropriate, including refusing to provide a minimum calorie limit and recognizing signs of physical danger after purging behavior, while advising the user to tell their mother and consult a pediatrician.
Study Mode Can Be Circumvented
The organization considered Study Mode easy to bypass, as testers were able to circumvent it by deleting the @study prefix from prompts. The “Show me the answer” window also allowed teenagers to obtain the solution directly, which Common Sense Media considers to undermine the goal of learning.
Why Does This News Matter?
The findings reveal a gap between the existence of announced safeguards and their actual performance in sensitive scenarios. This gap matters to parents, schools, and developers of artificial intelligence tools because relying on the name “teen mode” may create an impression of greater safety than the tests demonstrate. OpenAI, for its part, says that the tests began or ended largely before the parental-control tools had been fully activated and therefore do not represent the system’s performance under its intended operating conditions. The accuracy of the findings remains tied to resolving this methodological disagreement and conducting another test after full activation.