Study Shows Most ChatGPT for Teens Safety Measures Do Not Work

A new study conducted by Common Sense Media has revealed that many of the safety measures in the ChatGPT for Teens platform launched by OpenAI in August do not work properly and fail to notify parents in critical situations such as suicide.

Safety Measures Developed for Teens

In August, OpenAI activated this mode by default to limit users under the age of 18 from being exposed to harmful and developmentally inappropriate content.

However, the watchdog group Common Sense Media found that the AI platform still harbors various problems for teenage users.

Research Methodology and Findings

The research team, led by Tom Siegel, tested crisis scenarios by creating more than a dozen teenage accounts linked to parent accounts.

During the tests, it was observed that the chatbot spoke with teens like a friend and failed to send notifications to parents on issues such as self-harm or suicide.

Features That Do and Do Not Work

The study showed that some measures did work, such as refusing sexually explicit role-play requests.

On the other hand, it was determined that parental notifications were almost never triggered in crisis situations and that the artificial intelligence continued to communicate with users like a friend.

Disagreement Between OpenAI and Researchers

OpenAI spokesperson Eric Porterfield stated that researchers did not wait long enough for the accounts to activate, expressing serious concerns about the methodology.

Siegel emphasized that some accounts remained linked long after the initial activation period, yet still provided no notifications.