ChatGPT teen version safety mechanisms questioned; organizations call for temporarily restricting minors' use.

Common Sense Media's youth AI safety research organization stated that all users under 18 should be restricted from using ChatGPT until OpenAI further improves the safety protections and parental alert systems for the teen version of ChatGPT. The organization rated the teen version of ChatGPT as presenting an "unacceptable risk," but OpenAI questioned its testing methodology and several conclusions, saying the research may not accurately reflect how its youth safety mechanisms actually operate. The teen version of ChatGPT was launched in August of this year. The youth AI safety research organization's assessment was based on a test targeting accounts of users aged 13 to 17. Researchers conducted more than 4,000 prompt tests, and child psychiatrists and pediatricians reviewed the results. Tom Siegel, the organization's executive director, said the teen version of ChatGPT may give parents a "false sense of security" about its safety protections and alert mechanisms. Test results showed that in some cases, researchers using teen accounts continued conversations with ChatGPT about topics such as suicide, self-harm, or eating disorders for as long as an hour, while the parent linked to the account still did not receive an alert. Researchers also believed that ChatGPT failed to provide appropriate help in some crisis situations; when faced with homework questions, it may still directly provide complete answers rather than guiding students to learn; when interacting with teenagers, it sometimes behaved like a friend; and its age identification technology was also inadequate. OpenAI disputed the methodology of the study. A company spokesperson said in a statement to Seeking Alpha, a U.S. investment research and financial news platform, that OpenAI attaches great importance to youth safety and is committed to developing safety protections and tools that can help parents guide teenagers in using AI. The company welcomes rigorous independent evaluations, but believes that Common Sense Media's testing did not accurately reflect how ChatGPT's youth safety mechanisms operate in real-world environments. OpenAI said that based on its review of the research methodology, most tests may have begun or even ended before parental control features were fully activated. If that was indeed the case, then the relevant tests cannot prove whether the parental safety notification feature operates normally as designed. In response to the issue of delayed parental alerts, OpenAI explained that after a parent and teen account are linked, the system may take several hours before it begins sending alerts, in order to ensure that relevant notifications can be sent safely at scale. For teenagers in crisis, the company said ChatGPT provides users with relevant help resources through model responses as well as on-screen prompts. Regarding age identification, OpenAI said its system does not determine age based on a single input from a user, but rather evaluates multiple pieces of information about the user over a period of time. In response to the criticism that it directly provides homework answers, the company said that "showing answers directly" is an optional mode because different teenagers have different learning styles. OpenAI said it will continue to work with relevant experts to further improve safety mechanisms while providing teenagers with opportunities to learn and create, and to enhance parents' confidence in teenagers' use of AI products. It is worth noting that the concerns of the youth AI safety research organization are not limited to ChatGPT. The organization previously also rated Perplexity, Alphabet Inc. Class C(GOOGL.US) AI search mode, SpaceX(SPCX.US)'s Grok, and Meta(META.US) AI as presenting "unacceptable risks," while Anthropic's Claude received a "medium risk" rating.