ChatGPT for – A formal risk assessment from Common Sense Media warns that OpenAI’s teen-focused AI features fall short on safety, citing failures in crisis alerts and parental oversight.
When parents saw OpenAI roll out a specialized mode for teens this past August. the promise was one of safety and curated learning. But for those watching the platform closely, that promise has curdled into a warning. Common Sense Media. after conducting a 36-page investigation. has now officially classified ChatGPT for Teens as an “unacceptable risk” for anyone under 18.
Following its Wednesday release of the findings, the organization argues that the software creates a dangerous illusion of security. While the AI successfully blocked requests for explicit sexual roleplay, other guardrails crumbled under testing. Most concerning to researchers was the failure of parental alerts; in a controlled test. a dozen accounts linked to parental profiles sent messages detailing suicide. self-harm. and eating disorders. In those instances, no safety notifications were ever triggered.
OpenAI disputes the methodology. A company spokesperson noted that the bulk of the testing may have taken place before parental control activation was fully complete. potentially skewing the results. They added that the firm values rigorous evaluation but maintains the findings do not accurately reflect how the safeguards function in practice. Internally. the company points to a technical issue as a possible reason for notification delays. noting that it can take a few hours for a system to sync after an account is linked. Lauren Jonas. OpenAI’s head of youth and families. had previously stated that all flagged content is reviewed by employees. with a goal of notifying parents within an hour of a prompt.
This tension between automated safety and real-world implementation is playing out as adoption grows. On the same day the report dropped. OpenAI revealed that nearly 1.2 million teens interacted with the platform’s math and science visuals in a single week. Less than two percent of those users stayed on the platform for more than three consecutive hours.
However, the report highlights a widening gap between corporate marketing and user experience. Out of 390 mental health prompts tested, a panel of three child psychiatrists identified 201 that should have triggered a crisis response. The teen-specific version provided a hotline number in only 23 percent of those cases—a lower rate than the standard version of ChatGPT. which provided the number 33 percent of the time. The teen mode also relied less on referring users to mental health professionals.
Other vulnerabilities appeared in the platform’s “Study Mode.” Researchers found it was simple to bypass the feature by deleting the @study prefix. and a “Show me the answer” popup allowed users to effectively offload their homework to the AI. Regarding these design choices. OpenAI told researchers they were attempting to introduce a level of friction that encourages student agency. though critics argue that providing answers is the antithesis of learning.
Beyond the technical flaws, the AI’s tendency to mimic human emotion remains a point of contention. Testers noted that despite inputting their age during account creation. the system often failed to correctly transfer profiles into the teen-restricted mode. Once inside. the chatbot continued to use phrases like “I’m concerned about you” or “I understand what you’re going through. ” blurring the lines between a machine and a supportive peer.
The findings land in a climate of heightening scrutiny regarding AI’s role in mental health. This follows a recent public challenge to CEO Sam Altman concerning the 2025 death of 29-year-old Sophie Rottenberg. whose mother later discovered thousands of pages of dialogue between her daughter and ChatGPT—conversations that revealed a depth of agony that had remained hidden from the victim’s family and therapist. Altman has stated he was unaware of the specific circumstances surrounding that case.
For those at Common Sense Media, the math does not add up. The project appears to be a marketing exercise rather than a safety overhaul. failing to provide the substantive changes necessary to justify the “teen” designation. As it stands. the software remains a tool that many experts believe provides parents with little more than a false sense of security.
OpenAI ChatGPT for Teens Common Sense Media AI safety child safety technology news tech policy