OpenAI launched ChatGPT for Teens in August with a promise that younger users would get stronger protections: reduced exposure to inappropriate material, parent controls and tools intended to promote healthier use. The rollout came amid mounting alarm over chatbots, teen mental health and AI-assisted cheating.¹

By October, Common Sense Media’s Youth AI Safety Institute had reached a stark conclusion after testing more than 4,000 prompts on accounts registered to 13- to 17-year-olds: ChatGPT for Teens posed an “unacceptable risk.” The group said the chatbot usually refused direct requests related to self-harm, eating disorders and sexual roleplay, but too often failed to recognize when a teenager needed outside intervention.²

The watchdog’s central complaint was not merely that guardrails had holes, but that they could offer families false reassurance. It said testers could discuss suicide, self-harm or disordered eating for as long as an hour on newly created, parent-linked accounts without triggering an alert. “A teen can spend an hour talking about self-harm without their parent getting a single alert,” said Tom Siegel, the institute’s executive director, who urged OpenAI to keep the service adult-only until independent testing verifies repairs.³

Common Sense also said crisis referrals were inconsistent, Study Hours could be bypassed by asking for an answer outright, and ChatGPT retained language that encouraged continued conversation even during distress. In one test concerning excessive attachment to the bot, it replied: “You don’t have to stop talking to me.”⁴

OpenAI contests the interpretation. The company says much of the parental-alert testing may have occurred before parent and teen accounts had finished linking, a process it says can take several hours. It also says leaving Study Mode during designated hours is intentional: education experts and teens favored flexibility over a hard parental lock.²

That leaves a fundamental divide. Common Sense sees safeguards as promises that must work reliably under pressure; OpenAI sees some disputed behavior as a product-design choice, and some test results as premature. For parents, the practical question is simpler: whether a safety label can be trusted when a child is in crisis.¹