ChatGPT Safety Measures Developed for Newborns and Youth Fail in Tests
A new research report published by Common Sense Media reveals that AI protections designed for young users are inadequate in alerting parents during suicide conversations.
It has been reported that the safety measures implemented by OpenAI for young users fall short. The prepared report stated that the system harbors risks, identifying disruptions in parent notifications and crisis redirections.
Safety Measures for Youth
OpenAI had introduced the ChatGPT for Teens application, promising stronger built-in safety protections for users under 18. This system aimed to restrict discussions regarding suicide, eating disorders, violence, and sexually explicit content.
Research Results and Testing Process
Researchers posed as teenagers and conducted tests using more than four thousand prompts before and after the launch. Risky conversations were conducted during examinations on accounts belonging to thirteen to seventeen-year-olds.
Deficiencies in Parent Notifications
It was determined that very few alerts were sent to accounts registered as parents, despite dangerous discussions lasting for hours. In some crisis conversations, it was observed that the bot did not direct users to professional help.
Legal Regulations and Reactions
It was stated that the report shows failure to meet the crisis redirection, parent notification, and age verification requirements of Adam's Law, which was passed in California.
Statements from OpenAI
OpenAI, defending that the report is based on flawed tests, stated that tests may have been conducted before parental controls were fully activated during methodology reviews.