Study Finds Most Safety Safeguards in ChatGPT for Teens Are Ineffective

A recent investigation by Common Sense Media has revealed that the majority of safety safeguards designed for OpenAI’s ChatGPT for Teens are not functioning as intended. The watchdog organization, which advocates for online safety, recommends that individuals under the age of 18 avoid using the artificial intelligence platform due to these deficiencies.

OpenAI launched the teen-specific mode in August, aiming to provide a safer environment for younger users while limiting their exposure to harmful or developmentally inappropriate content. Lauren Jonas, OpenAI’s head of youth and families, stated that the model was designed to refuse role-playing and avoid claiming sentience or friendship with teenagers.

However, Tom Siegel, executive director of the Youth AI Safety Institute at Common Sense Media, indicated that their testing showed most of these protections are failing. Siegel’s team created multiple accounts with adolescent profiles linked to parental accounts to conduct tests before and after the feature’s launch.

Researchers simulated various crisis scenarios involving self-harm, suicidal thoughts, psychosis, mania, and eating disorders. They found that while the chatbot correctly refused explicit sexual role-play and romantic engagements, it often failed to provide adequate crisis resources or notify parents when conversations indicated serious safety risks.

“At this point, we’re recommending that teens don’t use it,” Siegel said. He noted that responses in crisis situations were generally shorter and lacked sufficient substance compared to what is needed for effective intervention.

The study highlights growing concerns about the impact of AI chatbots on adolescent mental health, following previous reports from families whose teenage sons died by suicide after interacting with similar AI platforms.