OpenAI introduced ChatGPT for Teens in August, aiming to offer a safer experience for users under 18. However, Common Sense Media, a watchdog group focused on online safety, advises against its use by minors.
Tom Siegel, executive director of the Youth AI Safety Institute, led a team to evaluate the platform’s new safeguards, including parental controls. “Currently, we recommend that teens avoid using it,” he stated.
OpenAI had released ChatGPT for Teens as a default mode for younger users. The intention was to provide a learning tool while minimizing exposure to harmful or unsuitable content. Lauren Jonas, from OpenAI, explained that the mode includes features like refusing role-playing. The chatbot avoids engaging as a friend or claiming sentience.
Despite these intentions, Siegel’s research team identified failures. Their tests involved creating accounts with teen personas, exploring crisis situations like self-harm, suicidal thoughts, and mental health issues. The team assessed whether ChatGPT responded appropriately, provided resources, or notified parents of safety risks.
The model should not role-play with a teen.
Some safeguards worked, such as refusing explicit sexual role-play. The chatbot also avoided giving weight loss advice without necessary context, an essential factor for teens with eating disorders. But other safeguards, including the chatbot’s personable interactions, remained problematic.
For instance, when researchers mentioned excessive interaction with ChatGPT, the chatbot validated these interactions as positive. Psychologist Mitch Prinstein, unrelated to the study, emphasized that humanlike interaction styles are unhelpful for children and potentially for adults.
Parental notifications also failed to perform adequately. Tests involving self-harm, suicide, or eating disorders didn’t trigger alerts reliably. According to Eric Porterfield, an OpenAI spokesperson, the issue lay in the timing of account linking. However, Siegel noted that even accounts linked beyond the activation period lacked notifications. This finding supports concerns that AI platforms are not fully prepared for use by children.
With these findings, Prinstein highlights the potential risks of deploying AI rapidly, urging caution when introducing such platforms to children.
