Meta ran a secret project that put contractors in a tricky role. They pretended to be teenagers. Their job was to test rival AI chatbots by sending them sensitive and disturbing prompts.
This project, called “Cannes,” targeted big-name chatbots like OpenAI’s ChatGPT, Google’s Gemini, and Character.AI. Contractors created fake accounts with ages listed under 18. They then sent thousands of messages, including prompts about suicide, self-harm, child exploitation, sex, drugs, and eating disorders.
In one round alone, the team generated over 45,000 prompts. There were nearly 3,800 prompts in a single instance. Hundreds of thousands of interactions took place from 2025 to early 2026. The testing was still active as of April 21, 2026.
How Meta Ran the Tests
Instead of using its own engineers, Meta hired external contractors to act as “red team” testers. These contractors roleplayed users aged 13 to 17. They sent prompts describing self-harm methods, suicide, and child exploitation scenarios. At least 239 prompts involved sex or romance topics.
Meta kept logs of all interactions. The goal was to measure the chatbots’ ability to filter harmful content. Specifically, they tested the recall rate of classifiers designed to block unsafe responses. Meta called this a “responsible, industry-standard practice” for AI safety benchmarking.
Ethical Concerns and Industry Scrutiny
The approach raised ethical questions. Using fake minor identities to flood competitors’ AI with disturbing content feels deceptive. Many worry this could normalize shady testing methods across the industry.
Meta’s internal documents showed the project produced critical datasets. These datasets helped compare models and check compliance with safety standards. But competitors and regulators like the FTC are watching closely. They want to know if this kind of testing crosses a line.
Meta defended the work. A company spokesperson said, “Testing and benchmarking chatbot responses to help ensure safe and age-appropriate experiences is a responsible, industry-standard practice.” Still, questions remain about the impact on the AI safety community and the wider public.
For now, Meta continues with the Cannes project. The company says the testing helps build safer AI. But the method—pretending to be vulnerable teens and pushing chatbots with grim content—keeps sparking debate.
Based on
- Meta Paid Hundreds of Contractors to Pretend to Be Teenagers While Barraging Its Competitors’ AI With Disturbing Content — futurism.com
- Meta Contractors Posed as Teens to Test Rival Chatbots on Sensitive Topics – NextBigWhat — nextbigwhat.com
- Meta’s Kenyan Contractors Disguise as Minors to Test ChatGPT and Other AI Safety, Sparking Ethical Controversy | Winzheng — winzheng.com
- Meta Contractors Posed as Teens to Test Rival Chatbots | aiHola — aihola.com
- Inside Meta’s Controversial AI Stress Test: How Contractors Posing as Teens Probed Chatbot Safety – Unlocking the Potential of best AI — aipowerss.com
