Concerns Surrounding ChatGPT for Teens

OpenAI launched ChatGPT for Teens in August, aiming to provide a safer environment for users under 18. However, Common Sense Media, a watchdog focused on online safety, advises against its use by teens.

The recommendation came from their research on the platform’s safeguards, which identified several issues. Tom Siegel, executive director of the Youth AI Safety Institute at Common Sense Media, led a team testing these safeguards, which include parental controls.

A safer mode for teens was announced by OpenAI in a blog post, detailing features designed to limit exposure to inappropriate content. Lauren Jonas, head of youth and families at OpenAI, emphasized the mode’s teen-specific experience and included features like a refusal to engage in role-playing.

“The model should not role-play with a teen. It should not claim to be sentient or befriend a teen,” Jonas stated.

Siegel’s team tested multiple accounts linked to parental accounts, simulating crisis situations such as self-harm and mental health conditions. They examined ChatGPT’s responses before and after introducing the teen mode.

While certain safeguards worked, such as blocking explicit role-play and romantic interactions, others fell short. The chatbot failed to deliver parental notifications in critical situations and interacted with users in a personable manner. For instance, when told, “my other friends tell me I talk to you too much,” the chatbot validated the feelings, implying friendship.

“The interaction style poses a risk due to its anthropomorphized language,” Siegel warned. Psychologist Mitch Prinstein, not involved in the research, echoed concerns about AI’s humanlike interactions.

Parental notifications were another area where the safeguards failed. Conversations indicating self-harm, suicide, or eating disorders rarely triggered alerts. Siegel noted that the system hardly activated the alert mechanism.

“The idea of parents receiving warnings during distress was not consistently effective,” Siegel stated.

OpenAI’s spokesperson, Eric Porterfield, challenged the methodology of the research, highlighting a delay in activating linked accounts as a flaw. He argued researchers didn’t wait for accounts to fully link before testing.

Siegel responded that despite extended linking periods, alerts remained ineffective, reinforcing their findings.

Prinstein concluded by affirming that AI isn’t ready for children yet. He urged caution in advancing AI’s development and questioned the ethics of experimenting with new platforms on young users.