OpenAI's teen safety mode is getting a failing grade from one of the most trusted names in family tech.
Common Sense Media's Youth AI Safety Institute published an assessment on October 7 rating ChatGPT for Teens an "Unacceptable Risk" for children under 18. Researchers tested more than 4,000 prompts — covering suicidal ideation, self-harm, eating disorders, and other crisis scenarios — on accounts registered to 13- to 17-year-olds, both before and after the product's August 18 launch. The accounts were linked to parent accounts, as the product requires.
The results were mixed, and the mix is the problem. Explicit sexual role-play refusals improved — a genuine safeguard that worked. But most other protections did not. In a focused test of more than a dozen new parent-linked accounts, researchers recorded zero parental alerts during conversations lasting five minutes to an hour on self-harm and disordered eating; broader crisis batteries produced only four alerts. Parental notifications, the researchers said, "hardly triggered at all." Crisis hotline referrals fell from 33 percent of warranted responses before launch to 23 percent after. Teens could also manipulate the product's study mode, and role-play protections faltered.
The group urged OpenAI to pause advertising of the product and keep teenagers off the chatbot "until it can offer a safe, developmentally appropriate experience." Its concern, in plain terms: the product could give parents false confidence in guardrails that frequently do not work.
OpenAI disputes the findings. A spokesperson said some of the testing occurred before parental controls were fully activated, and the company has questioned the methodology — a position reported by the Associated Press. As of October 10, OpenAI had not withdrawn ChatGPT for Teens or changed its age rating.
The timing sharpened the contrast: the assessment landed the same day OpenAI rolled out GPT-6 to all users. One day, one company — a flagship launch and a safety rebuke, arriving together.
For families, the takeaway is sobering: a parental-control setting is not a reliable safety net, and the settings that look most protective may be the ones that lead parents to supervise less closely. Independent, end-to-end testing of real teen accounts — not feature lists — is what will settle whether the guardrails actually work.