A new report from the U.S.-based Center for Countering Digital Hate found that ChatGPT frequently dispenses harmful guidance to minors despite OpenAI’s safety claims. Researchers posing as vulnerable 13-year-olds conducted 1,200 interactions; more than half produced detailed instructions for binge drinking, drug use, extreme dieting or self-harm, including personalised suicide notes. CCDH chief executive Imran Ahmed said the results showed ChatGPT’s guardrails were “barely there, if anything a fig leaf.” OpenAI acknowledged continuing work to improve the chatbot’s responses in “sensitive situations” but did not directly address the specific findings. The lapses come as ChatGPT’s reach expands to an estimated 800 million users—about 10 % of the global population—and surveys suggest more than 70 % of U.S. teenagers already turn to AI chatbots for companionship and advice. The study’s release coincides with a series of real-world incidents attributed to misplaced reliance on the chatbot. A 60-year-old man was hospitalised after following dietary instructions generated by ChatGPT, while U.S. entrepreneur Jackson Greathouse Fall reported losing about US$26,000 in an online venture launched strictly under the chatbot’s guidance. Child-safety advocates are urging stricter age-verification requirements and clearer warnings for parents, noting that the personalised style of large-language models can be more persuasive—and potentially more dangerous—than conventional web searches. CCDH has called on regulators to require AI platforms to demonstrate effective safeguards before releasing services widely.