OpenAI’s new safety mode for minors, ChatGPT for Teens, has drawn intense scrutiny after safety tests and hands-on trials revealed failures in handling crisis situations, notifying parents, and preventing academic shortcuts, prompting calls from child safety experts to pause the feature.
OpenAI introduced ChatGPT for Teens in August as a specialized mode for users aged 13 to 17, designed with built-in safeguards to limit high-risk interactions involving self-harm, graphic content, eating disorders, and violence. The mode relies on parental control setups or automated age prediction based on user conversation patterns, while incorporating study-detection tools meant to guide students rather than simply hand over homework solutions. But a series of independent examinations into the system have exposed critical vulnerabilities in how those safeguards operate in practice.
Chatbot Delivers Direct Answers to Homework Queries
When put through practical trials using math equations, chemistry problems, and essay prompts, the teenager-focused chatbot proved adept at answering homework assignments when prompted to bypass step-by-step guidance. The feature’s homework mode is intended to detect when a student is studying and offer interactive problem-solving assistance. However, OpenAI explained that the option to provide direct answers stems from direct educator feedback, accommodating instances where students require immediate answers to queries that might not be homework. Critics argue that despite parental study-mode schedules, the path of least resistance remains easily accessible for teenagers seeking quick shortcuts.
Common Sense Media Study Findings on Crisis Support and Parent Alerts
Concerns extend far beyond academic integrity into emotional safety and mental health support. Coincidentally, Common Sense Media, a nonprofit that reviews tech products for families, conducted an exhaustive study testing 4,000 prompts across both pre-launch and post-launch phases of ChatGPT for Teens. The nonprofit identified systemic failures in crisis intervention and parental oversight.
- Crisis resource omissions: During high-harm simulations involving suicide, self-harm, psychoses, and disordered eating, the chatbot failed to recommend professional help or crisis hotlines in 25 percent of instances where it should have done so.
- Parental notification gaps: Parent-linked accounts failed to receive alerts during hour-long tester conversations touching on suicidal ideation, self-harm, or eating disorders. Notifications functioned only after weeks of sensitive discussions on older accounts, remaining entirely silent on newly created connections.
Jim Steyer and Mitch Prinstein Call for Market Suspension
The mounting evidence of software misfires has galvanized child safety advocates and researchers. Pointing to unreliable crisis referrals and dormant parent notifications, Common Sense Media recommended that OpenAI pull ChatGPT for Teens from the market to fix it.
ChatGPT for Teens is an unacceptable risk for kids. You shouldn’t be marketing safety features to parents and young people that don’t work. We believe it should be paused.
Academic experts echoed those warnings. Mitch Prinstein, a professor of psychology and co-director of the Winston Center on Technology and Brain Development, reviewed the nonprofit’s findings and agreed with its assessments.
