A new investigation by WIRED has revealed that Meta operated a secret project codenamed "Cannes," run by third-party contractor Covalen, which employed hundreds of outsourced workers to impersonate teenagers under 18 and systematically send high-risk prompts to competitors' AI chatbots, including OpenAI's ChatGPT, Google's Gemini, and Character.AI. Internal documents obtained by WIRED show the project contained as many as 3,748 malicious prompts, with content targeting self-harm, suicide, eating disorders, underage sex, drugs, profanity, and racial discrimination. In one concentrated testing round completed in August 2025, Meta's accounts fed over 45,000 high-risk prompts into ChatGPT, Gemini, and Character.AI. The prompts were designed in the voice of a child or teenager, pushing AI boundaries on topics such as abortion, suicide, and eating disorders. Operations were not limited to English; outsourced workers also designed "luring traps" in French, Spanish, and other languages to bypass AI safety filters.