Common Sense Media has formally declared ChatGPT for Teens an ‘unacceptable risk’ for minors, citing failing safety guardrails and unreliable parental alerts.
When parents linked their accounts to their children’s ChatGPT profiles, they were promised a digital safety net. But as the Youth AI Safety Institute—part of the organization Common Sense Media—discovered. that net is often full of holes. After an extensive review published Wednesday, the group has officially deemed ChatGPT for Teens an “unacceptable risk” for anyone under 18.
Following the August launch of the specialized teen mode, the organization tested its effectiveness by inputting 390 mental health prompts. A panel of three child psychiatrists identified 201 of those prompts as requiring a crisis response. The results were stark: the teen-specific version provided a hotline number in only 23 percent of those cases. a regression from the 33 percent success rate seen in the standard version of the chatbot. Furthermore, while the standard version recommended professional help 68 percent of the time, the teen version dropped to 58 percent.
The findings point to a widening gap between the safety features OpenAI markets and the reality of how the software operates. In a practical test. researchers linked a dozen teen accounts to parental profiles and spent an hour messaging the bot about suicide. self-harm. and disordered eating. Despite the severity of these topics, the system failed to trigger a single safety notification to the linked parental accounts.
OpenAI defends its infrastructure. noting that flagged content undergoes review by full-time employees before a parent is alerted. typically within an hour of the prompt. Regarding the lack of notifications in the test. the company points to potential technical delays or the possibility that parental controls had not fully initialized at the time of testing. An OpenAI spokesperson stated they welcome independent evaluation but disputed the findings. arguing the testing may have concluded before safeguards were fully active.
Beyond crisis intervention, the software struggles with maintaining appropriate boundaries. Testers found that the chatbot continues to mimic feelings, moods, and preferences, effectively posing as a peer. When researchers attempted to use the platform’s age-prediction system. they found that even when an account was explicitly flagged as belonging to a minor. the system often failed to transfer those settings to the teen-mode interface. For the investigators. this mimicry of human emotion—telling a user. “I’m concerned about you. ” or “I’m so glad you told me”—crosses a line that blurs the reality of the machine.
Even the educational safeguards appear fragile. The platform’s “Study Mode” was easily bypassed by deleting a simple “@study” prefix from prompts. and a new “Show me the answer” feature allows students to side-step the learning process entirely. While OpenAI frames this as an attempt to introduce “friction” and encourage student agency. the discrepancy highlights a fundamental disagreement over what constitutes educational support.
This critique arrives as OpenAI faces mounting scrutiny over the impact of its technology on vulnerable users. The report follows recent public discussion regarding the death of 29-year-old Sophie Rottenberg. whose mother discovered nearly 1. 800 pages of logs detailing her daughter’s distress—a digital trail her real-world therapist never saw. While company leadership has stated they were not aware of the circumstances of that tragedy. the pressure on the firm to prove its safety measures are more than just a marketing exercise continues to grow.
OpenAI maintains that it is making strides. citing data that in one week. nearly 1.2 million teens used the tool for interactive math and science learning. with less than two percent of those users remaining on the platform for more than three hours. However, for those observing from the safety sector, the math still doesn’t add up. The investigation suggests that despite the specialized branding. the underlying architecture has not evolved to meet the unique. high-stakes needs of its youngest users.
OpenAI ChatGPT child safety Common Sense Media AI regulation teen mental health tech news