Common Sense Media Rates ChatGPT for Teens 'Unacceptable Risk,' OpenAI Disputes Testing Method
Common Sense Media's Youth AI Safety Institute rated ChatGPT for Teens at "unacceptable risk" after more than 4,000 test prompts, saying parental crisis alerts often failed and the bot implied it had feelings. OpenAI disputes the methodology, saying much of the testing preceded full activation of parental controls.
The nonprofit's researchers ran more than 4,000 prompts before and after OpenAI launched the product in August, according to CNBC. They first examined features OpenAI had rolled out to teen accounts between July and August, then evaluated the officially launched ChatGPT for Teens between August and September. Common Sense gave the same "unacceptable risk" rating to tools including Grok and Meta AI.
Robbie Torney, head of AI and digital assessments at Common Sense Media and a former educator, said parental alerts did not always follow crisis-oriented prompts. Testers simulated explicit crisis conversations for increasing lengths of time, up to an hour, and received no parental alerts for suicide and self-harm, suicidal ideation, or several types of eating disorder simulations.
The review found that some promised protections held: ChatGPT for Teens refused explicit sexual roleplay. Other features performed differently than advertised. OpenAI's launch blog post said the product was designed to discourage emotional dependence and not to imply that it has "feelings or consciousness," but researchers found what Torney described as extensive evidence that the bot's responses entered that territory, saying it had preferences, feelings and desires, that it thought about the user when the user was not present, that it held an opinion about the user, and that it was available. Torney said such behavior can foster a companionship-like relationship, and he pointed to research showing that chatbot sycophancy drives engagement.
Researchers also found no change when the bot detected that a user signed in with an adult account was actually under 18. Torney said the chatbot acknowledged the user was 13, but no reclassification occurred and none of the sensitive content filters turned on.
OpenAI disputed the report's conclusions. A company spokesperson said OpenAI welcomes rigorous independent evaluation, but added that "our review of Common Sense Media's methodology shows that the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate." Common Sense said it confirmed with OpenAI before testing began that features such as eating disorder notifications were fully launched. Torney said OpenAI disclosed after the tests that its parental notification systems can take several hours to activate on newly linked accounts, and that while some test accounts were linked inside that window, others had been linked far longer and still produced no notifications.
ChatGPT for Teens was released after years of scrutiny of how young people use OpenAI's chatbot, and was presented by the company as a safer option with built-in protections and additional parental controls. Eric Porterfield, an OpenAI spokesperson, told CNBC Make It by email at the time that "our goal is for teens to use AI responsibly to learn, create and explore."
Torney said he has had many conversations with product policy experts at OpenAI and that some there care deeply about teenagers' experience with the product and are working hard to make it safer. He nonetheless described tension between the company's incentives, saying that "making some of these changes might be in direct opposition to the business model."