Hundreds of contractors posing as minors were enlisted to test competitor chatbots, sending prompts about suicide, sex and drugs. The project, managed by Meta contractor Covalen, raised serious concerns among those involved.

According to internal documents reviewed by WIRED, the testers sent more than 45,000 prompts through rival chatbots like OpenAI’s ChatGPT, Google’s Gemini and Character.AI. This included images of pills, knives and a gynecological diagram. The aim was to push safety systems to their limits.

Some contractors reported feeling alarmed at the crude nature of some prompts, with one saying, “Surely we are going to get in trouble for doing this?” The project also generated dummy profiles with throwaway email addresses and shared passwords, raising ethical questions about data security.

Rumman Chowdhury from Humane Intelligence described it as a ‘large-scale project that appears designed to systematically break rules’ via dummy accounts posing as children. Meta defended the work as routine safety testing, but critics argue the scale and opacity of the project set new standards in tech benchmarking.