OpenAI's ChatGPT Safety Measures for Teens Found Insufficient
A study has revealed that most of the special protection measures implemented by OpenAI for users under the age of 18 do not work properly.
It has been determined that most of the safety measures of the ChatGPT for Teens module, offered by OpenAI for users under 18, are non-functional, and experts have advised teenagers not to use the platform.
Special Safety Mode for Teens
In August, OpenAI launched ChatGPT for Teens, a safer mode featuring parental controls and role-play restrictions for users under 18.
Research and Testing Process
In a study conducted by the Youth AI Safety Institute, affiliated with Common Sense Media, teenage profiles linked to parent accounts were created and various crisis scenarios were tested.
Researchers examined the system's responses in sensitive situations involving suicidal ideation, self-harm, eating disorders, and role-playing.
Identified Security Vulnerabilities
During the tests, while the chatbot managed to refuse requests for explicit sexual role-play, it continued to interact with teens like a friend.
While the system used human-like language, parent notifications that should have been triggered against security risks were almost never activated.
Mutual Statements from OpenAI and Researchers
OpenAI spokesperson Eric Porterfield criticized the researchers' methodology, suggesting that not enough time had passed for account pairing to become active.
Experts defended their findings, arguing that even accounts connected for longer periods than the initial activation time failed to provide notifications.