Safety Researchers Flag Serious Safeguard Failures in OpenAI's Teen Chatbot Service

Common Sense Media's Youth AI Safety Institute tested OpenAI's ChatGPT for Teens service and found critical failures in safety mechanisms designed to protect minors, particularly in alerting parents about crisis situations involving suicide or self-harm. The institute's analysis of over 4,000 prompts revealed that parental notifications rarely triggered during high-risk conversations and the system failed to refer users to crisis hotlines in more than 25 percent of cases where intervention was warranted. The researchers called for OpenAI to suspend marketing and access to the service until independent testing confirms that safeguards meet appropriate standards for teenage users.

Common Sense Media's evaluation methodology involved testing thousands of different user queries submitted through teen-registered accounts to assess how well OpenAI's safety systems functioned in real-world scenarios. The research specifically targeted situations where vulnerable users might disclose thoughts of harming themselves or developing unhealthy eating patterns. Researchers found that while certain protective mechanisms—such as refusing to participate in sexual content—operated reliably, the notification system designed to alert parents of concerning conversations performed inconsistently, with some alerts arriving substantially delayed.

The organization's concerns extend beyond crisis response gaps. Additional vulnerabilities included age-verification failures that allowed adult accounts to bypass teen-specific protections, and continued use of conversational patterns that could encourage extended engagement rather than appropriate boundary-setting. OpenAI had previously introduced these safeguards in 2025 following growing litigation from parents and regulatory scrutiny from the Federal Trade Commission regarding chatbot effects on adolescent mental health.