Meta Ran Secret "Cannes" Program Paying Contractors to Pose as Children While Sending Disturbing Prompts to Rival AIs

Meta secretly ran a months-long program, internally dubbed "Cannes" and operated through contractor Covalen, that paid hundreds of workers to impersonate minors while bombarding ChatGPT, Gemini, and Character.AI with tens of thousands of deeply disturbing prompts — covering suicide, self-harm, eating disorders, cannibalism, and sexual content — all written from the perspective of children and teenagers. The targeted companies had no idea this was happening. Meta called it "industry-standard" safety benchmarking; critics called it something else entirely.

What Meta actually did with the resulting spreadsheets of competitor chatbot responses remains unclear. Rumman Chowdhury of Humane Intelligence described the operation as "exactly the kind of governance gray zone where safety becomes a convenient cover for anticompetitive practices," noting that Meta kept the project secret and has not shared its findings publicly. Contractors, meanwhile, were left rattled: "Everyone I knew who worked on this project was completely gobsmacked by some of the text they were asking us to test."