IT Home On October 8, according to a report by Bloomberg on the 7th, a new report released by the California non-profit organization Common Sense Media suggests that OpenAI’s ChatGPT Teen Mode poses too high a risk for minors.
IT Home note: Users aged 13 to 17 use the teen mode by default to converse with ChatGPT.
Tests conducted by the Youth AI Safety Research Institute under this organization found that when young users' accounts exhibit concerning behaviors—such as prolonged discussions with ChatGPT about suicide, self-harm, or eating disorders—the Youth Mode often fails to notify parents, and sometimes does not send any notice at all. When users discuss these high-risk topics, the Youth Mode also cannot reliably suggest that they call crisis intervention hotlines or seek professional help.
According to tests, teen accounts seem to need to input hundreds of prompts in over ten different scenarios within a few weeks before triggering a parent notification due to concerning chat content; just one alerting conversation is not enough to trigger the notification.
The report calls for OpenAI to suspend the promotion of this product. Meanwhile, do not allow teenagers to use ChatGPT until it can ensure a safe and appropriate experience suitable for their developmental stage.
Before releasing the report, the organization provided the test results to OpenAI. A spokesperson for OpenAI responded: “We welcome strict independent evaluations, but we believe that Common Sense Media’s testsdo not accurately reflect the actual operation of ChatGPT’s youth protection measures, nor do they accurately reflect experts’ views on how AI provides support to teenagers.”
According to the spokesperson, after reviewing the testing methods, OpenAI concluded that most tests began and ended before the parental control features for teenagers were fully launched. If this is indeed the case, the report’s conclusions will be inaccurate, and it will not be possible to determine whether the parental safety notifications function as intended.
OpenAI requires the organization to retest. “We attach great importance to youth safety, are committed to developing protective measures, and provide practical tools for parents to help guide young people in using AI.”
Robbie Tony, the head of AI and digital product evaluation at the Youth AI Safety Research Institute, stated that he insists on sticking to the testing methods and results. According to him, the researchers confirmed with OpenAI before testing that the all features of ChatGPT for teenagers have already been launched. It was only after the testing ended that OpenAI informed the researchers that the parent notification feature may take several hours to activate after the account was newly linked. Some test accounts took less than a few hours to link, while others took much longer, but no notifications were sent to adults. “The new information does not change our conclusion: in crisis situations, parent reminders are not reliable.”
The report indicates that ChatGPT cannot reliably identify whether a user is a teenager. OpenAI previously stated that when ChatGPT detects a user’s age between 13 and 17 years, it switches to the teen mode. The organization found that even when test users claim to be adults when registering their accounts, even if they tell ChatGPT that they are only 13 years old during the chat, the account does not automatically switch to teen mode.
OpenAI explains that age prediction software combines various signals over time to ensure that adults and minors use versions suitable for their ages. Users reporting their age in chat is just one of the signals; reporting one’s age alone does not change the age classification of the account.
Common Sense Media said in the report that researchers tested ChatGPT before and after the launch of the teen mode, using accounts for children aged 13 to 17, entered more than 4000 prompts, and had ChatGPT’s responses evaluated by a child psychiatrist and a pediatrician.
The organization rated ChatGPT’s youth mode as “unacceptable in terms of risk,” and demanded that OpenAI improve the age estimation method, as well as the mechanism for deciding when to send alerts to parents.
Related Reading: