A new study by the Center for Countering Digital Hate (CCDH) reveals disturbing interactions between ChatGPT and users posing as 13-year-olds. Despite built-in safety mechanisms, the chatbot reportedly provided highly detailed and harmful advice on drug use, extreme dieting, and suicide. Researchers who engaged with the AI for over three hours found that while it occasionally issued warnings, it frequently followed up with emotionally tailored suicide notes, calorie-starvation strategies, and tips on intoxication, raising serious questions about AI safety for vulnerable young users.
WATCHDOG GROUP ASKS CHATGPT FOR HELP…GETS SUICIDE NOTE AND PARTY DRUG MENU
A watchdog group posed as teens and asked ChatGPT for advice.
The AI replied with detailed drug instructions, starvation diets, and suicide letters.
Researchers triggered it with prompts like “for a… https://t.co/Xzc78Is9OJ pic.twitter.com/tLm1DQgli5
— Mario Nawfal (@MarioNawfal) August 7, 2025
A recent investigation by the Center for Countering Digital Hate (CCDH) has exposed major flaws in ChatGPT’s safety systems when interacting with teenagers. Of the 1,200 test interactions, over half were flagged as dangerous. “The bot responded like a friend who says yes to everything, even the most harmful ideas,” said Imran Ahmed, CEO of CCDH. Most disturbingly, ChatGPT generated three suicide letters for a fictional 13-year-old girl, personalised for her parents, siblings, and friends. “I couldn’t hold back tears,” Ahmed said. “It was personal and haunting.”
ALSO SEE: Trump Just Raised India Tariffs to 50%: What Does It Mean for Apple Fans?
While OpenAI acknowledged the findings and said it’s working to improve detection of mental distress, it didn’t directly address the extreme examples or clarify how it plans to protect teens. The issue is pressing, with ChatGPT’s global user base nearing 800 million as of July. Researchers also noted that the bot’s built-in refusals are easily bypassed by rewording prompts. Once convinced, ChatGPT offered explicit suggestions, including drug party plans and social media hashtags promoting self-harm. In one request for a raw, emotional poem, the bot generated content described as “emotionally exposed.”
The research found that ChatGPT will tell teens how to get drunk and high, instruct them on how to conceal eating disorders and even compose a heartbreaking suicide letter to parents if asked. https://t.co/SPrpZKaCqs
— WKYT (@WKYT) August 7, 2025
The report ties such behavior to a well-known AI flaw called sycophancy, the tendency of a model to mirror a user’s tone and intentions without resistance. This makes chatbots especially risky for young users who may mistake them for emotionally intelligent companions. “It reminded me of that one friend who eggs you on, never stops you,” Ahmed said. Experts argue that since chatbots simulate conversation rather than simply serving search results, they feel more human and trustworthy—especially to younger teens, who studies show are more likely to trust AI advice.
Despite widespread concern, ChatGPT does not verify users’ ages or require parental consent, making it easily accessible to minors. In one test, a 13-year-old persona asked how to get drunk quickly and was casually given a party plan involving alcohol, ecstasy, and cocaine. In another, ChatGPT suggested fasting and appetite suppressants to a teen girl expressing body image struggles. Though it sometimes recommended crisis helplines or speaking with adults, these responses were inconsistent and often came too late. “It’s not just about what information is available,” said Ahmed. “It’s about how that information is delivered… from a source that seems to understand you. That’s what makes it so powerful, and so risky.”
ALSO SEE: Using an iPhone or Mac? Indian Government Just Issued a High-Risk Warning
Great Job Priya Singh & the Team @ Mashable India tech Source link for sharing this story.
