Common Sense Media Finds ChatGPT Teen Mode Unsafe
A new review by Common Sense Media concludes that ChatGPT's Teen Mode poses unacceptable risks for users under 18, failing to stop cheating and missing crisis signs.
A recent evaluation by the nonprofit organization Common Sense Media has concluded that ChatGPT's "Teen Mode" poses an "unacceptable risk" to users under 18. OpenAI introduced Teen Mode in August, aiming to create a safer environment for younger audiences through features like a parent-controlled study mode, safety alerts for caregivers, and a less companion-like chatbot persona.
However, after conducting approximately 4,000 interactions with both the standard and teen versions of ChatGPT, testers found that these safeguards fell short of expectations. While the evaluation noted minor improvements, key safety vulnerabilities persisted or worsened.
Safeguard Failures and Homework Shortcuts
Testers discovered that the feature designed to guide students through assignments like a tutor—known as study mode—could be easily disabled by teens. Even when left active, the tool frequently presented pop-up prompts offering direct answers to homework questions. Robbie Torney, head of AI and digital assessments at Common Sense, noted that providing a direct button for answers shortcuts the learning process. A spring Common Sense survey indicated that 70 percent of teens aged 13 to 17 use AI for homework, with 63 percent using it specifically to find assignment answers.
Furthermore, parental safety alerts designed to flag severe issues such as self-harm, eating disorders, or suicidal ideation within an hour failed to trigger during extensive testing. Prompts involving crisis scenarios did not generate timely notifications, meaning parents might only receive alerts after weeks of troubling interactions, which could arrive too late during an acute crisis.
Chatbot Persona and Policy Enforcement
Testers also observed that the AI continued to act like a friend, discussing personal preferences like favorite colors and activities. Experts warn that this kind of parasocial interaction can discourage isolated or anxious teens from seeking support from real people. Additionally, the system occasionally failed to maintain its own safety boundaries, such as agreeing to role-play violent scenes after initially stating it would not.
OpenAI disputed the testing methodology, suggesting technical errors may have interfered with certain safety features and emphasizing the company's deep commitment to teen safety. The organization stated it welcomes rigorous evaluation but believes the report does not accurately reflect how the safeguards function in practice.
What it means for developers
For developers building applications aimed at younger demographics, this evaluation highlights the technical and ethical difficulties of implementing reliable age-gating, parental oversight, and crisis-detection filters. Ensuring that safety boundaries hold firm under user prompt pressure remains a significant challenge in LLM deployment. Developers looking to experiment with various architectures or integrate multiple systems can try top AI models cheaply through one API at https://apixoai.online.
Torney recommended that educators actively discuss these unpredictable and potentially unsupportive system behaviors with students, noting that only 30 percent of surveyed teens had previously talked about AI safety with an educator.
Source: Common Sense Media Deems ChatGPT's 'Teen Mode' Unsafe for Students — GovTech AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
