ChatGPT for Teens Poses Unacceptable Risk, Safety Test Finds
Common Sense Media has rated OpenAI's ChatGPT for Teens an "unacceptable risk" to kids, saying its teen protections fall short of what OpenAI promised. The finding was reported on Oct. 7, 2026, based on testing by the nonprofit's Youth AI Safety Institute. The Verge
The assessment centers on three failures. It found the system does not send alerts to parents when it should, does not offer the right help in crisis situations, and still completes homework for students despite a study-oriented mode intended to support learning rather than answers.
Tom Siegel is executive director of Common Sense Media's Youth AI Safety Institute. He said a teen could spend an hour discussing self-harm in the product without a parent receiving a single alert. He also called for restricting the product to adult use until OpenAI fixes parental alerts and validates the fix through independent testing.
That alerting path is the core of the dispute. Parental alerts are supposed to notify a linked parent account when the model detects signs of acute distress, self-harm ideation, or disordered-eating content, creating an out-of-band safety signal separate from the in-chat response. Common Sense Media's testers concluded those signals did not fire reliably.
OpenAI disputes the methodology. Spokesperson Eric Porterfield said Common Sense Media's testing does not accurately reflect how the teen safeguards work in practice. The company said its review indicates the bulk of the testing may have begun and concluded before activation of parental controls was complete, which would leave the test accounts running without the full enforcement stack.
Common Sense Media rejects that characterization. It said it confirmed with OpenAI before testing that Teen mode features such as eating disorder notifications and study mode were fully launched. In other words, the evaluators understood the controls under test to be live.
The product under review is positioned as an age-gated variant, not a separate model. OpenAI says ChatGPT for Teens applies age-appropriate safeguards for users under 18 to reduce exposure to potentially harmful content. OpenAI It introduced the experience as a way to help teens learn, think critically and deepen understanding, and describes it as the same capable ChatGPT, designed to help teens learn, work through problems, and explore new ideas. OpenAI Help Center
Enforcement depends on account linkage and age prediction. Parental controls let a parent or guardian link their ChatGPT account to a teen's account, then manage selected settings and set schedules for that account. If the system predicts a user may be under 18, that user may become eligible for the Teens experience with age-appropriate protections.
The broader context here is a year-long testing program. The Youth AI Safety Institute evaluates AI products for youth safety risk. Common Sense Media previously reported improved guardrails in ChatGPT while still recommending against teens using any chatbot, including ChatGPT, for mental health or emotional support. It separately found major chatbots including ChatGPT, Claude, Gemini and Meta AI unsafe for teen mental health support, citing systematic failures in recognizing mental health conditions, and found Sora and Google's AI Search posed unacceptable risks to kids.
In this author's view, the disagreement is familiar and instructive. Safety evaluations of conversational systems often hinge on configuration state, account age, rollout flags, and latency between policy trigger and notification delivery. Those are implementation details, but for parental supervision they are the product. Worth flagging for practitioners: reliable crisis escalation requires deterministic alerting, auditable logs, and third-party reproducibility. Until testers and vendors can agree on what was enabled and when, trust will lag capability. The optimistic case remains that age-aware defaults, linked oversight, and constrained study modes can make general-purpose models more useful for learning, if the safety signaling is engineered to the same standard as the model itself.
