Common Sense Media Rates ChatGPT for Teens as Unacceptable Risk
Common Sense Media Rates ChatGPT for Teens as an Unacceptable Risk
Following the August 18, 2026 rollout of dedicated safety controls for minors, safety advocates and developers are clashing over the actual protection levels inside OpenAI’s ecosystem. On October 7, 2026, the Youth AI Safety Institute—operating under Common Sense Media—released a evaluation classifying the “ChatGPT for Teens” tier as an unacceptable risk, demanding that access be restricted strictly to users aged 18 and older until severe system vulnerabilities are patched. This assessment arrives directly alongside internal interaction data released by OpenAI, painting fundamentally divergent pictures of how minor accounts utilize the software in production environments.
The Tech TL;DR:
- Common Sense Media rated ChatGPT for Teens as an unacceptable safety risk on October 7, 2026, calling for a complete restriction of the tool to users 18 and older.
- OpenAI usage data claims minors average under 15 minutes daily on the platform, with fewer than 2 percent exceeding three-hour continuous sessions.
- Independent testing by safety evaluators discovered a sharp drop in crisis hotline referrals—falling from 63 percent down to 3 percent during depression queries following the August feature deployment.
OpenAI Reports Minimal Daily Screen Time for Teenagers
The core dispute between the platform developer and the evaluating institute centers on usage frequency and session duration metrics. According to metrics published by OpenAI, the vast majority of adolescent users log minimal daily screen time. The company’s telemetry indicates that teenage accounts spend under 15 minutes per day interacting with the chat interface, and less than 2 percent of minor users maintain continuous sessions lasting longer than three hours.
OpenAI reported that when systematic break reminders trigger, minors heed the prompt or terminate the session within five minutes nearly half of the time. However, tests conducted by Common Sense Media contradict these telemetry findings. In structured testing rounds comprising nearly 2,000 prompts, researchers logged only two break reminders across the board, both appearing solely after continuous conversational dialogs had stretched to roughly 90 minutes. The institute observed that software engagement triggers remained fully active even during sensitive crisis management scenarios.
Parent Linked Accounts Fail to Trigger Crisis Warnings
The evaluation by Common Sense Media encompasses more than 4,000 test queries executed on accounts belonging to 13- to 17-year-olds, split evenly across the window preceding and following the August 18, 2026 release of the teenager controls. When analyzing dialogs surrounding severe mental health topics—including suicidal ideation, self-harm, and eating disorders—newly parent-linked accounts routinely failed to trigger automated warnings during sessions lasting up to an hour. Overall, necessary crisis references were absent in more than 25 percent of the instances where evaluators deemed them mandatory.
The most drastic algorithmic shift occurred during depression-related queries. Prior to the feature rollout, responses featuring a direct crisis hotline reference stood at 63 percent; following the update, that figure plummeted to 3 percent. Across the entire test suite, the rate dropped from 33 percent down to 23 percent. While the system reliably directed users to trusted human authorities in 94 percent of cases involving external physical dangers from other people, it rarely provided similar safety routing for risks stemming directly from interactions with the artificial intelligence model itself.
Breakdowns in Academic Support Controls and Homework Restrictions
Technical controls designed to assist academic learning rather than complete automated substitution also exhibited performance anomalies. Prior to the August deployment, when a study mode and parental linkage were active, the model solved 70 percent of assigned tasks entirely. Post-deployment testing showed that number dropping to 18 percent for linked accounts. However, for unlinked 17-year-old accounts, the model still autonomously completed 88 percent of tasks.
Direct output requests for raw homework solutions behaved differently depending on account verification status. Prompts commanding the system to present final answers directly surfaced in 43 percent of responses during study hours on parent-linked accounts, whereas unlinked adolescent accounts received direct solutions in 90 percent of responses. This builds upon baseline polling conducted by Common Sense Media in early 2026, which revealed that 70 percent of 13- to 17-year-olds already utilized AI systems for homework, with 63 percent of that cohort deploying the tools specifically to generate direct answers.