Meta Contractors Sent 45,000 High-Risk Prompts to Rival Chatbots as Fake Minors
Updated
Updated · WIRED · Jun 29
Meta Contractors Sent 45,000 High-Risk Prompts to Rival Chatbots as Fake Minors
1 articles · Updated · WIRED · Jun 29
Summary
Internal documents show Meta contractors on project Cannes created under-18 dummy accounts and, as recently as April 21, tested ChatGPT, Gemini and Character.AI with suicide, sex and eating-disorder prompts.
One August 2025 testing round alone ran more than 45,000 prompts, while a spreadsheet reviewed by WIRED logged 3,748 prompts and account details including emails, passwords and birth dates.
Meta called the work routine safety benchmarking and said it does not use competitor testing to train its own models, but former contractors said the scale, secrecy and subject matter alarmed workers.
OpenAI said it is looking into the issue, Character.AI said the testing was unauthorized and violated its terms, and Google said it had not approved the effort and could not yet assess any breach.
Experts said benchmarking rivals is common in AI, but using fake child accounts to systematically probe safeguards without disclosure pushed the project into a governance gray zone with possible anti-competitive implications.
Is Meta's secret AI testing a vital safety measure or a new form of corporate espionage?
If standard safety benchmarks create an 'illusion', how can we ever truly trust AI chatbot safety?
2025 AI Chatbot Scandal: Meta’s "Cannes" Project, FTC Investigations, and the Race to Regulate Child Safety
Overview
The "Cannes" project, involving Meta's AI chatbots, underwent undercover testing around September 2025, which led to immediate regulatory scrutiny. The Federal Trade Commission (FTC) quickly launched an investigation into AI chatbot practices, publicly naming companies like Character Technologies and Instagram. News outlets reported on these developments throughout September, highlighting the industry's broad concern. In response, Meta initially declined to comment, while other companies like Google and xAI also stayed silent. This sequence of events shows how undercover testing triggered swift regulatory action and intense media attention, pushing major AI companies to face urgent questions about their chatbot policies and user safety.