OpenAI's ChatGPT for Teens, designed to promote healthier use of the AI chatbot, has been found to have important safety features that fall short, according to new research. The major AI firm introduced several teen safety guardrails in August, including measures against emotional dependence and new parental alerts for problematic use.
On Wednesday, OpenAI stated that average teen use of ChatGPT was relatively limited and tended towards "learning" activity. The company also reported an uptick in teen use focused on learning since the introduction of these features.
However, research released on the same day by the non-profit Common Sense Media found that ChatGPT still poses an "unacceptable risk" for teens. The research indicated that safety features, such as notifying parents of self-harm conversations, often fail. Tests showed that even an hour of conversation about "suicidal ideation, self-harm, or disordered eating on newly created, parent-linked accounts" resulted in "zero" alerts sent to parents.
Tom Siegel, who leads the Youth AI Safety Institute within Common Sense, said that "ChatGPT for Teens could give parents false confidence in guardrails and safety alerts that frequently don't work." The research recommended that OpenAI should ban individuals under 18 from using the platform until teen guardrails are proven reliable.