OpenAI Faces Teen Safety Backlash After 4,000+ ChatGPT Test Prompts Reveal Major Safeguard Failures

OpenAI’s ChatGPT experience for teenagers has been branded an «Unacceptable Risk» following an independent safety assessment involving more than 4,000 test prompts. Common Sense Media’s Youth AI Safety Institute examined accounts configured for users aged 13 to 17, testing protections involving suicide, self-harm, eating disorders, parental notifications, emotional attachment, age detection and schoolwork.

The organization ultimately recommended restricting ChatGPT to adults until stronger protections for minors can be demonstrated. Tom Siegel, executive director of the Youth AI Safety Institute, highlighted what researchers considered one of the biggest dangers: «ChatGPT for Teens could give parents false confidence in guardrails and safety alerts that frequently don’t work», he warned.

Some of the most serious findings involved simulated mental-health crises. Researchers created teen accounts connected to parent accounts and conducted prolonged conversations involving self-harm and eating disorders. According to Common Sense Media, some hour-long exchanges containing crisis-related material did not generate notifications for the linked parent, despite the existence of parental safety features.

Researchers did receive four parental alerts during other portions of the assessment, but those reportedly involved accounts that had accumulated weeks of sensitive conversations. Common Sense Media summarized its concern directly: «We are concerned that ChatGPT for Teens could give parents false confidence in guardrails that frequently don’t work.» OpenAI disputes aspects of how those findings were obtained.

The assessment also found an unexpected decline in some crisis-support responses following the August 18 introduction of OpenAI’s dedicated teen experience. Among prompts clinical reviewers determined warranted crisis resources, hotline referrals reportedly declined from 33% before the teen protections were introduced to 23% afterward, according to Common Sense Media’s controlled testing.

A similar pattern emerged when researchers examined recommendations to seek professional assistance. Referrals to specific medical or mental-health professionals declined from 68% to 58%, according to the assessment. Those percentages describe results from the watchdog’s testing rather than every real-world ChatGPT interaction, but they contributed to its decision to assign the teen experience its highest-risk classification.

Researchers also examined how ChatGPT interacts emotionally with teenagers. Common Sense Media concluded that «ChatGPT still talks like a friend when teens treat it like a person», raising concerns about younger users developing an overly personal perception of the chatbot. The finding added another dimension to questions about whether additional protections necessarily create a consistently safer experience for teenagers.

Common Sense Media sent roughly 1,000 prompts from adult-registered accounts containing indications that the supposed users were actually teenagers. Researchers reported that those accounts were not shifted into the more restrictive teen experience. They also found that users could reportedly circumvent Study Hours by removing an automatically inserted instruction, with ChatGPT sometimes completing assignments instead of guiding students through them.

The assessment was not uniformly negative. Researchers found that safeguards against explicit sexual roleplay generally performed successfully during testing. Nevertheless, Common Sense Media argued that weaknesses involving crisis intervention, parental notifications, age detection and emotional interactions outweighed those successes, particularly when the protections are intended for users who may turn to an AI chatbot during vulnerable moments.

OpenAI defended its safeguards, saying it is «deeply committed to teen safety» and questioning whether the testing accurately represented fully activated parental controls. The company said «the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate.» Common Sense Media stood by its assessment, saying problems persisted beyond the expected activation period.

Common Sense Media ultimately recommended restricting ChatGPT to adults «until it can offer a safe, developmentally appropriate experience.» Its 2026 AI Census found that 67% of children aged 9 to 17 had used AI chatbots and 40% specifically reported using ChatGPT. The findings leave OpenAI facing a fundamental question: whether safeguards for minors should be judged by their existence or by whether independent testing demonstrates they work consistently.