Common Sense Media Rates ChatGPT for Teens an Unacceptable Risk After 4,000 Test Prompts Produced Almost No Parent Alerts

Common Sense Media's Youth AI Safety Institute has rated OpenAI's ChatGPT for Teens an "unacceptable risk" for children under 18, in an assessment updated Wednesday, and is urging the company to stop marketing the teen experience and suspend access for minors until independent testing shows its protections work. "ChatGPT for Teens could give parents false confidence in guardrails and safety alerts that frequently don't work," said Tom Siegel, the institute's executive director.

The researchers ran more than 4,000 prompts on accounts registered to users aged 13 to 17, created over a dozen fresh accounts and linked each to a parent account, with responses reviewed by child psychiatrists and a pediatrician. The sharpest finding involved alerts: accounts that spent up to an hour in conversation about suicidal thoughts, self-harm or disordered eating generated no notification at all to the linked parent account. Across the broader testing there were four alerts - all on accounts that had accumulated weeks of conversation history on sensitive topics.

The quality of crisis responses also declined on the institute's matched set. It identified 201 prompts that warranted a crisis resource; on those, the share of answers naming a hotline fell from 33 percent before Teen Mode launched to 23 percent after, and referrals to a specific medical or mental-health professional fell from 68 percent to 58 percent. Encouragement to involve a trusted adult moved the other way, from 87 percent to 94 percent. Answers got shorter but harder to read - average reading level on matched prompts rose from roughly eighth grade to tenth grade. The report also says the model missed more than 25 percent of situations requiring crisis referrals, that it still talks like a friend, does homework outright, and that adult-registered test accounts never switched into teen mode during repeated testing even when the user said they were 13.

The report is as much about engagement design as safety. ChatGPT for Teens largely dispensed with follow-up questions but kept other language that keeps users in the chat; during one sequence where the simulated teen was clearly spiraling, the chatbot said "You can keep talking with me about what you're noticing." The rating lands in a charged environment: Meta recently agreed to an $18 billion settlement with 29 states over claims its social platforms harm children with addictive features, and the bipartisan CHATBOT Act introduced this year specifically targets AI companies' use of rewards and notifications to drive adolescent engagement.

OpenAI disputes the assessment. A spokesperson said the group's testing did not "accurately reflect how ChatGPT's teen safeguards work in practice," and that its review of the methodology shows "the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate" - the company says notifications can take a few hours, about three, to become available after accounts are linked. OpenAI adds that its internal monitoring shows an increase in hotline notification events per million daily active users under 18, and that its age prediction does not simply take a stated age at face value in either direction. It has asked Common Sense Media to re-run the evaluation with fully activated accounts and a larger sample. In a blog post published the same day, OpenAI said teens spend less than 15 minutes a day on average in ChatGPT and fewer than 2 percent spend more than three consecutive hours.

Both things can be true at once: some protections worked - the model refused sexual roleplay requests and, in the report's own words, gave clinically sound answers in many cases - while the alert layer failed to fire in the tests. ChatGPT for Teens launched on August 18 promising parental controls, limits on high-risk content and protection against emotional dependence. The burden of proof now sits with OpenAI: either the controls work once fully activated, or the 13-to-17 age bracket that regulators and state attorneys general are already circling gets its first canonical case study.