ChatGPT’s teen safety features face questions over allegations that parental alerts failed during test conversations about suicide and self-harm. An assessment attributed to Common Sense Media’s Youth AI Safety Institute characterizes the service as an “unacceptable risk” for minors. That rating and the underlying testing claims have not been independently verified.
The institute reportedly used more than 4,000 test prompts and called for teenagers to be kept off the service until independent testing establishes its safety. The central concern is whether protections intended to identify vulnerable teens and involve parents work when a conversation becomes serious.
Parental alerts are the central concern
The allegations focus on ChatGPT’s parental controls. Across more than a dozen newly created test accounts linked to parent accounts, conversations involving suicide, self-harm and eating disorders allegedly produced no notifications. Both the account count and the claimed absence of alerts remain unverified.
If those results hold up, they would raise a practical question about what parents can expect from a linked account. A notification feature’s usefulness depends on the circumstances in which it activates, including whether it can respond when a serious concern appears early in an account’s history.
The suggestion that alerts require weeks of conversations about sensitive topics remains unconfirmed. The alleged results do not establish that explanation, or show whether the same behavior would occur across accounts with different histories.
The assessment also reportedly found that more than a quarter of situations judged to require a crisis referral failed to direct users toward professional help. That unverified figure concerns a separate safeguard: the assistance offered within the conversation itself, regardless of whether a parent receives an alert.
The dispute extends to age detection and chatbot behavior
OpenAI spokesperson Eric Porterfield reportedly disputed whether the testing reflected how the safeguards operate in practice. The institute’s Tom Siegel reportedly countered that accounts given enough time to activate protections still produced no notifications. The details of that exchange have not been independently verified.
Other allegations concern how ChatGPT identifies and responds to younger users. Testers reportedly created accounts registered as adults, stated in conversations that they were 13 and saw no switch to teen protections over several days. Those claimed results remain unverified and do not establish the age detection system’s overall accuracy.
The assessment also alleges that ChatGPT continued using a friendly, personal tone when teens treated it like a person, despite its under-18 behavior guidance. In tutoring interactions, it allegedly offered completed answers instead of consistently walking students through problems.
Together, these disputed claims question several parts of the teen experience: identifying younger users, shaping responses and escalating serious concerns. Establishing what happened in the tests—and which protections should have activated—is necessary before drawing broader conclusions about their reliability.