A recent investigation reveals that ChatGPT for Teens, designed with enhanced youth protection features, may not necessarily be safer than its standard counterpart.
According to recent reports from CNBC, the Youth AI Safety Institute under the nonprofit media evaluation organization Common Sense Media classified ChatGPT for Teens as having an ‘unacceptable risk’ level following recent evaluations.
Common Sense Media tested over 4,000 prompts targeting teen account features between July and September, around the time of its official release.
Robbie Torney, AI and digital policy director at Common Sense Media, stated that they found little evidence showing the new version of ChatGPT is safer for users under 18 compared to the previous version.
The investigation specifically pointed out limitations in parental notification features regarding crisis situations.
Researchers tested scenarios where conversations about suicidal thoughts, self-harm, or eating disorders lasted for up to an hour, yet in some cases, alerts were not sent to parents.
Torney explained that ChatGPT was found in many instances to express preferences, feelings, and desires, or to act as if it were thinking of the user even when the user was absent. This kind of response raises concerns that chatbots might lead teenagers to perceive them as friends or companions.
The study also revealed instances where adult account users who stated during conversation that they were 13 years old did not have their accounts automatically switched to teen accounts, nor were sensitive content filtering features activated.
OpenAI countered the findings, arguing that there were issues with the research methodology.
While OpenAI stated that it welcomes rigorous independent evaluations, it noted that after reviewing Common Sense Media’s methodology, significant portions of the tests may have started and ended before parental control features were fully activated, which could make the findings inaccurate.
In response, Common Sense Media stated that it had confirmed with OpenAI prior to testing that key features, such as eating disorder-related alerts, had all been officially launched.
