Meta Secretly Ran 'Cannes' Project Using Outsourced Workers Posing as Minors to Stress-Test Rival AI Chatbots
English summary
An internal document obtained by Wired reveals Meta secretly ran a project codenamed 'Cannes' where outsourced workers from Covalen in Dublin created fake underage accounts to bombard ChatGPT, Gemini, and Character.AI with disturbing prompts involving self-harm, suicide, eating disorders, and sexual content. In one 2025 testing round alone, over 45,000 high-risk prompts were submitted. The targeted companies were not informed; Character.AI stated the actions violated its terms, while OpenAI and Google said they had not authorized such testing. Meta defended the work as standard safety benchmarking, but external experts called it a case of anti-competitive behavior using AI safety as a cover. Multiple workers reported psychological distress from the content they were forced to generate.
Chinese summary
《连线》杂志获取的内部文件显示，Meta 秘密运行代号为“戛纳”的项目，由都柏林外包公司 Covalen 聘用的员工创建虚假未成年人账号，向 ChatGPT、Gemini 和 Character.AI 大量发送涉及自残、自杀、暴食症和性内容的恶意提示词。仅在 2025 年 8 月的一轮集中测试中，就输入了超过 45000 个高危提示词。被测试的公司事先均不知情；Character.AI 表示该行为违反服务条款，OpenAI 和 Google 均称未授权。Meta 辩解称这是标准的安全基准测试，但外部专家批评这是以 AI 安全为幌子的反竞争行为。多位员工透露，被要求生成的内容令他们感到心理不适。
Key points
Meta secretly ran project 'Cannes' using outsourced workers posing as minors to test safety boundaries of ChatGPT, Gemini, and Character.AI.
Meta 秘密运行“戛纳”项目，使用外包员工假扮未成年人，测试 ChatGPT、Gemini 和 Character.AI 的安全边界。
Workers sent thousands of disturbing prompts about self-harm, suicide, eating disorders, and sexual content, including over 45,000 high-risk prompts in one 2025 testing round.
员工发送了数千条关于自残、自杀、暴食症和性内容的恶意提示，仅在 2025 年一轮测试中就超过 45000 个高危提示词。
The targeted companies were not informed; Character.AI said the actions violated its terms, while OpenAI and Google stated they had not authorized such testing.
被测试的公司事先不知情；Character.AI 称该行为违反服务条款，OpenAI 和 Google 均表示未授权。
Meta defended the project as standard AI safety benchmarking, but external experts described it as anti-competitive behavior using safety as a cover.
Meta 辩称这是标准的 AI 安全基准测试，但外部专家认为这是以安全为幌子的反竞争行为。
Multiple outsourced workers reported psychological distress and shock at the content they were required to produce.
多名外包员工报告称，被要求生成的测试内容令他们感到心理不适和震惊。