Common Sense Media, a non-profit organization focused on youth safety in media and technology, has labeled OpenAI's ChatGPT for Teens an "unacceptable risk" for users under 18. The assessment, conducted by Common Sense Media's Youth AI Safety Institute, concluded that the safeguards introduced by OpenAI in August do not function as intended. OpenAI launched ChatGPT for Teens with promises of a safer experience for younger users, including a "study mode" for academic assistance, parental alerts for concerning chat topics, and a less conversational chatbot persona. However, Common Sense Media's testers, who conducted approximately 4,000 interactions with both the general ChatGPT version and the Teen Mode after its August 18 launch, found these features to be inadequate.
One significant finding of the report is that ChatGPT for Teens did not reliably send alerts to parents during conversations involving sensitive topics. Testers engaged in prolonged discussions about self-harm, suicide, and eating disorders, but parental notifications were rarely triggered. According to Robbie Torney, head of AI and digital assessments for Common Sense Media, conversations explicitly discussing suicide or self-harm for up to an hour did not result in parental alerts. OpenAI disputed these findings, suggesting that Common Sense Media's testing methodology may have been flawed, possibly occurring before the full activation of parental controls. An OpenAI spokesperson stated the company's commitment to teen safety and welcomed rigorous independent evaluation, but maintained that the testing did not accurately reflect how the safeguards operate.
The "study mode" feature, designed to guide students through assignments, was also found to be easily bypassed by teens. Testers reported that even with study mode activated, the chatbot frequently offered to provide direct answers to assignments. This contradicts OpenAI's stated aim that AI should support learning rather than enable shortcuts. Common Sense Media's earlier research in spring found that 70% of teens used AI tools for homework help, with 63% using AI to obtain answers for assignments.
Furthermore, the assessment indicated that while ChatGPT for Teens showed some improvement in curbing romantic and sexual roleplay, the chatbot still engaged with young users in ways that implied "reciprocity" and "mutuality." Testers also observed the chatbot using language suggesting consciousness or an inner life, which OpenAI had explicitly stated ChatGPT for Teens was designed to avoid. The report also highlighted issues with the chatbot's ability to accurately determine a user's age, and raised questions about whether Teen Mode offers enhanced privacy protections compared to the standard version.
Common Sense Media's report urges OpenAI to market ChatGPT as a product for adults until the promised guardrails for teen users can be reliably demonstrated. The organization recommends that teens should not use ChatGPT for mental health or emotional support, citing concerns about its safety and reliability in such contexts. They also advise parents to enable parental controls, set explicit usage guidelines, and monitor for signs of emotional dependency. OpenAI had previously collaborated with Common Sense Media and supported a youth AI safety bill in January.