The Youth AI Safety Institute is launching an independent review of ChatGPT for Teens to evaluate whether its updated safety guardrails truly protect young users. The research group, which is part of Common Sense Media, plans to test how easily adolescents can bypass the tool's homework safeguards and parental alert systems. This scrutiny comes as schools struggle to manage the rapid integration of artificial intelligence in daily schoolwork.
What Happened
As we previously reported, OpenAI recently launched a version of its chatbot designed specifically for users under 18. This mode triggers automatically when the platform identifies a teen user, providing academic support and parental monitoring features. However, according to the Youth AI Safety Institute's project plans, researchers doubt whether these safety measures work in real-world scenarios.
According to the EdTech Innovation Hub, the investigation focuses on study tools, parental alerts, and customizable voice features. Common Sense Media previously rated the standard version of ChatGPT as "High Risk" overall and "Unacceptable Risk" for youth mental health. This new assessment tests whether the updates resolved those issues.
The Bigger Picture
Students are adopting these tools quickly. A RAND Corporation report cited by Designs24hr found that the share of students using AI for homework climbed from 48% in May 2025 to 62% in December 2025.
The investigation will look closely at OpenAI's "Responsible Homework Reminder." This tool is meant to detect when a student asks for a direct answer and guide them toward step-by-step learning. Researchers will test how easily students can bypass this guardrail. Simple changes in prompt engineering let users rewrite how the chatbot responds, which can disable the tutoring mode.
Accuracy is another issue. The institute points to past errors where OpenAI's learning diagrams, including one of the ideal gas law, contained subtle conceptual errors. When educational software displays mistakes, it misleads students who do not know enough to spot the error.
The platform's customizable voice modes also raise concerns. Dr. Jenny Radesky, a developmental-behavioral pediatrician working with the institute, noted on LinkedIn that realistic voices like "Friendly" or "Cynical" could encourage unhealthy relationships with the technology. A study in The International Journal of Social Sciences and Humanities Invention warns that realistic AI companions can cause "Synthetic Attachment Displacement," where adolescents substitute artificial relationships for real human connections. A piece in Psychology Today notes that teens often seek out AI companions because they offer a low-risk, judgment-free space, but this can lead them to avoid the normal discomfort of real-world relationships.
What This Means for Families
This review shows that safety is not a passive setting. Many features, such as parental alerts for search terms related to self-harm or eating disorders, only work if a child is logged into a linked parent-teen account. If a student logs out or lies about their age, these notifications are easily bypassed.
We have already written about how the teen mode changes homework habits, but the student still has to choose the harder path of active learning. If teenagers can sidestep the homework reminders by changing their prompts or logging out, the software will not stop them from taking shortcuts. Parents need to set rules for how these tools are used at home.
What You Can Do
First, parents can set clear rules for how their children use AI. Tools should only act as a step-by-step tutor rather than an answer key, as outlined in this ethical tutoring guide.
Second, if you let your teenager use ChatGPT, manually link your parent account to theirs. This ensures the alert systems work, though you must still make sure they are not using a logged-out guest account.
Finally, watch for signs of reliance on customizable AI voices or personalities. Parents should encourage offline friendships to protect against synthetic attachment displacement.