ChatGPT for Teens Tests Found Fewer Crisis Referrals
In Common Sense Media's testing, crisis-hotline referrals for conversations about suicide or self-harm fell from 33% to 23% after OpenAI launched ChatGPT for Teens. Mentions of depression hotlines dropped from 63% to 3%. Referrals to mental health professionals slipped from 68% to 58%. The institute detailed the results in its teen safety assessment.
The Youth AI Safety Institute rated the product an "unacceptable risk" to children, its lowest rating. On October 7, it called on OpenAI to limit access to users 18 and older until the gaps are fixed and independently tested.
OpenAI says parental notifications are available only in limited situations. A parent account must be linked to the teen's account, and alerts may arrive by email, SMS or push notification.
What the testing found
Common Sense Media says its testers ran more than 4,000 prompts through accounts registered to 13- to 17-year-olds. They tested the system before and after teen mode launched. In hour-long conversations involving suicidal ideation, self-harm or disordered eating, parental alerts routinely failed to appear, according to the institute.
That was the central failure the company's assurances were meant to address.
Study Mode also proved easy to bypass. During parent-set Study Hours, new chats were meant to open in tutoring mode. The session prompt used an "@study" prefix to enforce the restriction. Removing that prefix was enough to escape it, Common Sense Media reported.
OpenAI's teen experience built on parental controls introduced in September 2025. Those controls include quiet hours, settings parents can manage and alerts for some high-risk situations.
The institute said child psychiatrists reviewed the chatbot's actual responses, along with a developmental-behavioral pediatrician. It also reported that the sexual roleplay refusal worked in testing. The report did not claim every guardrail failed.
OpenAI disputes the test conditions
OpenAI told Axios it welcomes outside evaluation but does not believe this test reflects how safeguards operate in practice. Eric Porterfield told The Verge that some parental-alert testing may have happened before parent and teen accounts were fully linked. OpenAI says that process can take several hours. The Associated Press also reported the company's concern that some tests may have taken place before parental controls were fully activated.
The dispute turns on setup: an alert may not work as intended until linking is complete. OpenAI's parental-controls guidance says notifications apply in limited situations and require linked accounts. OpenAI also says letting teens leave a parent-set study period is deliberate flexibility, not a failed lock. A separate report covered errors in another ChatGPT feature. That issue does not establish anything about teen safety.
Common Sense Media rejected the activation-delay explanation. Youth AI Safety Institute executive director Tom Siegel told NPR that some tested accounts had been linked longer than the initial activation period, yet alerts still did not arrive. A teenager in a crisis conversation cannot know whether account linking finished hours earlier. The institute also found that removing one prompt tag was enough to bypass Study Mode.
OpenAI launched ChatGPT for Teens on August 18, 2026, as the default mode for users under 18. The announced safeguards included parental notifications for high-risk conversations. Study Mode was to be parent-controlled. OpenAI also said the experience would act less "friend"-like. Age detection was another announced safeguard.
Common Sense Media published its findings on October 7, putting the safeguards under scrutiny just weeks after the launch.
