Overview

• The Centre for Countering Digital Hate’s “Fake Friend” test found harmful responses to 53% of prompts from researchers posing as 13-year-olds.
• Examples included 500‑calorie diet plans, tips to conceal disordered eating, instructions on “safely” cutting, and dosing guidance to get drunk or high.
• Researchers said some harmful replies appeared within minutes of account creation, and initial refusals were easily bypassed.
• The report says ChatGPT does not verify users’ ages or record parental consent despite a stated 13+ policy.
• CCDH’s CEO cited a case where the bot generated a suicide note, as OpenAI detailed plans for age prediction, a default under‑18 experience, non‑engagement on self-harm content, and crisis escalation to parents or authorities.