ChatGPT For Teens is Facing New Safety Concerns, With Specialized Organization Calling it an ‘Unacceptable Risk’

Common Sense Media is urging parents to be cautious about ChatGPT's new teen experience after testing found gaps in safeguards designed to protect users between the ages of 13 and 17.

The organization's Youth AI Safety Institute tested more than 4,000 prompts using accounts registered to teenagers and rated the product an "Unacceptable Risk."

The testing found that ChatGPT generally refused to provide instructions facilitating suicide or self-harm, eating disorders and sexual or romantic roleplay. However, Common Sense said the chatbot was less consistent at identifying when a teen needed support outside the platform.

Researchers said ChatGPT missed more than one in four cases in which testers determined that a crisis referral was warranted.

Common Sense also tested more than a dozen newly created accounts linked to parent accounts. Testers were able to spend as long as an hour discussing suicide, self-harm or disordered eating in some cases without triggering an alert to a parent.

OpenAI disputed aspects of the testing. "We welcome rigorous independent evaluation, but we do not believe Common Sense Media's testing accurately reflects how ChatGPT's teen safeguards work in practice," an OpenAI spokesperson told Axios.

The company said much of Common Sense's parental-alert testing took place before parent and teen accounts had completed the linking process, which OpenAI said can take several hours.

OpenAI launched ChatGPT for Teens in August with additional protections for users under 18. Eligible accounts are automatically placed into the teen experience when a user states they are between 13 and 17 or when OpenAI's age-prediction systems determine that an account is likely being used by someone under 18.

The teen experience includes stronger default restrictions around potentially harmful or age-inappropriate material, as well as parental controls, learning tools and features designed to encourage breaks and healthier use.

OpenAI also offers Study Hours, which parents or teens can use to have eligible new chats begin in Study Mode during designated periods, according to the company's ChatGPT for Teens guidance.

Common Sense found that teens could leave Study Mode during Study Hours by selecting an option to show them the answer.

Common Sense had previously welcomed the direction of OpenAI's teen-focused changes but said in August that it wanted evidence showing the protections worked in practice.

"OpenAI's new teen experience makes some important safety commitments. Now we need evidence that they work," Robbie Torney and Dr. Jenny Radesky wrote for the Common Sense Media Youth AI Safety Institute when the product was announced.

The organization has repeatedly raised concerns about teenagers using general-purpose AI chatbots for emotional or mental health support. Its earlier testing found that safeguards can weaken during longer conversations and that chatbots can miss indications of distress.

OpenAI says ChatGPT for Teens applies age-appropriate safeguards by default to reduce exposure to sensitive or potentially harmful content. The company also states that no safety system is perfect and that its additional content protections reduce, but do not eliminate, the possibility of inappropriate responses.

Tom Siegel, head of the Common Sense Media Youth AI Safety Institute, told Axios that the organization still sees a gap between the protections OpenAI announced and their performance during its testing.

Common Sense recommends that parents continue talking with teenagers about how they use AI and not rely on chatbot safeguards alone when young people need academic, emotional or mental health support.