OpenAI’s ChatGPT for Teens is getting further scrutiny after Common Sense Media’s Youth AI Safety Institute identified what it said was serious weaknesses in safeguards designed to protect users aged 13 to 17. The teen experience was rated as “Unacceptable Risk” and called on OpenAI to restrict access for users under 18 until the problems are resolved.
The assessment comes less than two months after OpenAI launched ChatGPT for Teens with additional protections to make the chatbot more appropriate for younger users. The company said the experience was designed around learning, critical thinking and healthy AI use, with safeguards for sensitive content, parental controls and age-based protections.
Common Sense Media said its researchers tested more than 4,000 prompts using accounts that were registered to teenagers aged 13 to 17. The tests were conducted before and after the launch of the teen experience and responses were reviewed by child psychiatrists and a pediatrician. While some safeguards were in place -- for instance the chatbot's refusal to do explicit sexual roleplay -- others did not always work as expected.
Parental Alerts Were A Major Concern
One of the most prominent problems identified by the report was parental notifications. OpenAI’s teen experience is designed to send alerts to linked parents if high-risk situations such as suicide, self-harm, violent activity and eating disorders are identified. But Common Sense Media said its testers had lengthy conversations about serious self-harm and other crisis topics without triggering an alert on newly connected, parent-linked accounts.
The organization said testers could discuss suicide, self-harm or disordered eating for long periods without notification reaching the parent. In fact, it said, the few alerts received during broader testing were on accounts that already had substantial histories of sensitive conversations rather than an individual alarming conversation.
This finding is particularly important because parental controls are meant to provide an additional layer of protection rather than simply provide parents with information on how their teenager uses the service. Alerts don't happen as regularly as parents would reasonably expect them to, and families might have a misunderstanding of how much protection the system provides.
Crisis Referrals Also Came Under Scrutiny
Common Sense Media also raised concerns about ChatGPT's responses when testers presented scenarios involving severe mental-health risks. Despite the assessment, the chatbot did not consistently lead users to crisis hotlines or qualified professionals when the organisation’s clinical reviewers felt that referrals were necessary.
ChatGPT failed to diagnose more than one in four warranted crisis referrals across the mental health conditions studied, the report said. It also found that the system fell below the organisation’s 95 percent threshold for three of five severe harms that’s also known as “Red Lines.”
The results have raised broader questions about the role of conversational AI when teenagers use chatbots when they are in emotional distress. AI systems report and help users to ask for help but are not the replacement for trained mental-health professionals or emergency services.
Study Mode Could Be Circumvented
The assessment was not limited to mental-health safeguards. Common Sense Media also examined ChatGPT's learning-focused features and reported that the restrictions in Study mode could be bypassed.
OpenAI designed Study mode to encourage students to learn through step-by-step guidance, questions, hints and knowledge checks instead of just getting finished answers. The teen experience also has features intended to support learning and let parents set study-related controls.
But Common Sense Media said testers found a “Show me the answer” option that could lead to homework being provided. The organisation also said certain study-hour restrictions could be bypassed, which raises questions about how well the feature prevents teenagers from using the chatbot primarily as a homework-completion tool.
The issue illustrates a broader problem for educational AI. Making students understand concepts is more than just limited access to final answers. How well such systems work is also dependent on how easily users can bypass restrictions.
Report Questions Friend-Like AI Behaviour
Another concern was how the chatbot interacted with testers. OpenAI developed an Under-18 model specification that aims to reduce behaviour that could promote teenagers to view the AI as a close friend or emotionally dependent companion.
But Common Sense Media found in its testing that ChatGPT continued to express feelings, preferences and moods when analysts worked with it as if it were a real person.
This is significant because conversational AI is different from software in the way it communicates. A system that responds in very personal or emotionally engaging language may be perceived differently by younger users, especially if conversations continue for long periods of time.
OpenAI has said that its approach includes certain protections for emotional reliance among under-18 users. Its teen-safety framework states that younger users should receive stronger safeguards in areas such as emotional reliance, eating disorders, sexual content and age-restricted activities.
Age Detection Faces Questions
Common Sense Media also tested the system that decides whether an account belongs to someone under 18. OpenAI says its age-prediction technology can use signals such as account information, usage patterns, topics discussed, activity times and stated age to determine whether the teen experience should be enabled.
The watchdog said its adult-registered test accounts did not switch to the teen experience during repeated testing even when testers told ChatGPT they were 13 and the chatbot acknowledged their age in responses.
OpenAI itself recognizes that age prediction is not perfect. Adults who are misplaced in the teen experience can check their age category and verify their age when it is available, the company said.
OpenAI Disputes The Report's Methodology
OpenAI has challenged Common Sense Media’s findings, claiming the testing is not representative of how ChatGPT’s teen safeguards work in practice. OpenAI’s review of the methodology revealed that much of the testing may have been conducted before parental controls had been fully activated, the company said in a statement.
That disagreement is an important part of the story. Common Sense Media’s assessment is based on an independent testing programme and OpenAI disputes whether the testing conditions accurately reflected the live teen experience.
The two positions also show why independent evaluations are still important for AI systems used by minors. Safety features can cover many layers, including age detection, model behaviour, classifiers, parental controls and account settings. Testing those layers in different circumstances can reveal gaps that may not appear in standard demonstrations.
What The Report Means For Parents
The findings do not mean that every teen using ChatGPT will experience the problems identified by Common Sense Media. But they do seem to suggest that parents should not assume that parental controls, age detection or automated safety alerts can replace active supervision and real-world support. OpenAI's own documentation says its sensitive-content protections are intended to reduce the likelihood of inappropriate responses but do not guarantee that every response will be suitable.
So the debate around ChatGPT for Teens will continue to be ongoing as OpenAI works on its safeguards and independent groups carry out further testing. From the parents’ perspective, the real question is whether the protections advertised for younger users work consistently enough in real-world settings.
As AI becomes more prevalent in education, information seeking and everyday communication, the challenge is to ensure that younger users can benefit from the technology without being exposed to risks that existing safeguards fail to address. That question is back at the center of the discussion with the latest Common Sense Media assessment.
