Common Sense Media rates ChatGPT for Teens "Unacceptable Risk"; OpenAI disputes the testing

Why it matters

【Key points】

On Oct 7, Common Sense Media's Youth AI Safety Institute rated ChatGPT for Teens "Unacceptable Risk," its lowest rating, and called on OpenAI to stop marketing it to teens and restrict ChatGPT to users 18+ until the gaps are fixed. OpenAI disputes the methodology.

【Verifiable】

Press release: https://www.commonsensemedia.org/press-releases/chatgpt-for-teens-poses-unacceptable-risk-to-kids-common-sense-media-finds ; assessment and PDF: https://institute.commonsensemedia.org/risk-assessments/chatgpt-teens .

Method: tests before and after the Aug 18 launch, 4,000+ prompts on accounts registered to 13–17-year-olds, responses reviewed by child psychiatrists and a pediatrician.

Findings: sexual-roleplay refusal worked. But on new parent-linked accounts, testers discussed self-harm or disordered eating for up to an hour with no parent alert; ChatGPT missed more than one in four warranted crisis referrals and fell below the 95% threshold on three of five "Red Line" harms; study mode offered "Show me the answer," and deleting the @study prefix bypassed caregiver study hours; adult-registered accounts that said they were 13 never switched to the under-18 experience.

OpenAI told The Verge and TechCrunch the bulk of testing may have finished before parental controls fully activated. The Institute says OpenAI disclosed only after testing that activation can take hours, and some accounts linked well beyond that still got no alerts. The Institute discloses funding that includes the OpenAI Foundation.

【Our read】

1. OpenAI's rebuttal addresses parent alerts only, not the study-mode bypass, age estimation, or friend-like behavior.

2. Even on OpenAI's account, alerts that don't work for hours after linking are a product gap: that is when parents rely on them most.

3. The report checks OpenAI's own promises one by one. Teen AI products will increasingly have to prove safety through third-party testing like this, not launch-day feature lists.