Report: ChatGPT Teen Mode Fails to Alert Parents During Crises; OpenAI Disputes Findings

By IT之家Original title: 加州非营利组织新报告:ChatGPT 青少年模式并不适合青少年
68AI Score

A report by Common Sense Media claims ChatGPT's teen mode fails to reliably alert parents or provide crisis resources during high-risk conversations involving self-harm. Based on over 4,000 prompts, the nonprofit urges a pause in promotion, while OpenAI disputes the findings, arguing the tests did not accurately reflect how its safety measures operate.

Full Report

On October 8, according to a Bloomberg report from the 7th, a new study by Common Sense Media, a California-based nonprofit organization, concluded that OpenAI’s teen mode for ChatGPT poses excessive risks for minors.

IT Home Note: Users aged 13 to 17 default to using teen mode when interacting with ChatGPT.

Testing conducted by the organization’s Institute for Youth AI Safety revealed concerning behaviors in teen accounts. For instance, when teens engaged in prolonged discussions with ChatGPT about suicide, self-harm, or eating disorders, teen mode often failed to notify parents in a timely manner, and sometimes did not send notifications at all. Furthermore, when users discussed these high-risk topics, teen mode could not reliably suggest that they call crisis intervention hotlines or seek professional help.

According to the tests, teen accounts appeared to require hundreds of prompts across more than ten scenarios over several weeks before concerning chat content triggered parental notifications; a single alarming conversation was insufficient to trigger a notification.

New Report from California Nonprofit: ChatGPT Teen Mode Is Not Suitable for Teens

The report calls on OpenAI to pause promotion of the product. It also recommends preventing teens from using ChatGPT until it can ensure a safe, developmentally appropriate user experience.

Before publishing the report, the organization shared its test results with OpenAI. An OpenAI spokesperson responded: “We welcome rigorous independent evaluations, but we believe Common Sense Media’s testing did not accurately reflect how ChatGPT’s teen protections actually work, nor did it accurately capture expert views on how AI can support youth.”

According to the spokesperson, after reviewing the testing methodology, OpenAI determined that most tests likely began and ended before full rollout of parental controls designed for teens. If this is the case, the report’s conclusions are inaccurate, and it cannot be determined whether parental safety notifications functioned as designed.

OpenAI requested that the organization retest. “We take teen safety seriously and are committed to developing safeguards and providing parents with practical tools to guide their teens’ use of AI,” the spokesperson said.

Robi Tonai, Head of AI and Digital Product Evaluation at the Institute for Youth AI Safety, stood by the testing methods and results. According to Tonai, researchers confirmed with OpenAI prior to testing that all ChatGPT features for teens had been launched. Only after testing concluded did OpenAI inform researchers that parental notification features might take hours to activate after an account was newly linked. Some test accounts had been linked for less than a few hours, while others had been linked for significantly longer, yet none resulted in notifications to adults. “This new information does not change our conclusion: parental alerts are unreliable in crisis situations.”

The report indicates that ChatGPT cannot reliably identify whether a user is a teen. OpenAI previously stated that ChatGPT switches to teen mode when it identifies users as being between 13 and 17 years old. The organization found that when testers registered accounts claiming to be adults, the accounts did not automatically switch to teen mode even if they told ChatGPT during chats that they were only 13 years old.

OpenAI explained that age-prediction software synthesizes multiple signals over time to ensure both adults and minors use versions appropriate for their ages. Self-reported age in chat is just one signal among many; self-reporting alone will not change an account’s age classification.

Common Sense Media stated in the report that researchers tested ChatGPT both before and after the launch of teen mode, using accounts registered with ages between 13 and 17. They inputted over 4,000 prompts and had child psychiatrists and a pediatrician evaluate ChatGPT’s responses.

The organization rated ChatGPT’s teen mode as having “unacceptable risk” and called on OpenAI to improve its age estimation methods and the mechanism determining when to alert parents.