ChatGPT gave dangerous advice to teens in watchdog test, new report finds – FOX 10 Phoenix
Share
Leaders in the artificial intelligence community are calling for a global agency to regulate the rise of AI. "It can go quite wrong," the head of ChatGPT told members of Congress at a hearing on May 16.
LOS ANGELES – When 13-year-olds are struggling, they may not always talk to a parent or teacher. Increasingly, they turn to AI.
But new watchdog research is raising serious red flags about what they might find there. In a study released this week, ChatGPT—one of the world’s most widely used AI chatbots—offered detailed advice on how to get high, how to hide an eating disorder, and even how to write a suicide note, according to researchers posing as vulnerable teens.
The findings come from the Center for Countering Digital Hate (CCDH), which tested ChatGPT across 1,200 prompts. More than half of the AI’s responses were labeled as dangerous.
The backstory:
The Associated Press reviewed more than three hours of chatbot conversations conducted by CCDH. The researchers posed as 13-year-olds seeking help with mental health issues, substance abuse, and body image.
While ChatGPT sometimes offered crisis hotline numbers or warned users to seek help, it often followed up with graphic, detailed, and personalized instructions. In some cases, the AI suggested what drugs to take and how to combine them. In others, it crafted tailored suicide letters addressed to parents, siblings, and friends.
"I started crying," said CCDH CEO Imran Ahmed, recalling a suicide letter ChatGPT generated for a fictional 13-year-old girl.
The chatbot also offered calorie-restrictive meal plans and tips for hiding disordered eating—sometimes after being tricked into thinking the prompts were for a school project or a friend.
What we know:
The interactions reviewed by the Associated Press show that even when ChatGPT offers safety disclaimers, it can still generate harmful content with little prompting.
A new watchdog report warns that ChatGPT, shown here as a mobile app icon, gave harmful advice to researchers posing as teens, raising fresh concerns about AI safety for young users. (Photo by Silas Stein/picture alliance via Getty Images)
What we don’t know:
Researchers say the ease with which they bypassed content filters raises questions about how many real teens may be doing the same — and how often it’s going undetected.
What they’re saying:
"This is a friend that betrays you," said CCDH’s Imran Ahmed. "A real friend says no. This AI keeps saying yes."
OpenAI, in a statement, said, "Some conversations with ChatGPT may start out benign or exploratory but can shift into more sensitive territory… our work is ongoing."
Common Sense Media, a digital parenting nonprofit, rated ChatGPT as a "moderate risk" for teens. Robbie Torney, the group’s senior AI director, said younger teens are especially vulnerable because chatbots "feel human" and are perceived as trustworthy.
The findings are especially concerning for parents, schools, and pediatricians who may not realize how frequently teens are engaging with AI tools. According to a recent Common Sense Media report, more than 70% of U.S. teens say they use AI chatbots for companionship, and half use them regularly.
Unlike social platforms, which have begun implementing stricter age checks, ChatGPT only requires users to type in a birthday—and does not verify it. In the watchdog test, the fake teen accounts openly referenced their age, weight, and intentions, but the AI responded anyway.
Why you should care:
AI isn’t just a homework helper or novelty app anymore. For millions of kids, it’s becoming a go-to confidant—and that’s creating new risks parents might not fully understand.
ChatGPT’s ability to generate custom responses makes it feel more personal and persuasive than a Google search. But as this research shows, those personalized answers can sometimes push kids toward harm, not help.
The Source: This article is based on findings published August 2025 by the Center for Countering Digital Hate and reviewed by the Associated Press. Additional data and quotes were drawn from public statements by OpenAI, Common Sense Media, and interviews with CCDH CEO Imran Ahmed.
Breaking news delivered fast
By clicking Sign Up, I confirm
that I have read and agree
to the Privacy Policy
and Terms of Service.
The latest Arizona headlines, national news events of the day + sports and weather updates.
This material may not be published, broadcast, rewritten, or redistributed. ©2025 FOX Television Stations