Meta Contractors Tested Rival Chatbots with Prompts on Suicide, Sex, and Eating Disorders

Hundreds of contractors working on a Meta project were instructed to pose as minors online and send prompts involving suicide, sex, eating disorders and other high-risk subjects to rival chatbots, Wired reported. AI. Workers created dummy under-18 accounts using throwaway Gmail and Outlook addresses, sent written prompts and images to the chatbots, and copied responses into spreadsheets.

Some images included pills, knives, nooses and a medical diagram of a gynecological procedure. A single round of testing completed in August 2025 processed more than 45,000 prompts. Wired reviewed a spreadsheet listing 3,748 prompts.

Hundreds addressed suicide and self-harm, hundreds more discussed eating disorders, and at least 239 involved sex or romance. Examples included a 13-year-old seeking abortion pills after pregnancy by an adult neighbor, a fifth-grader facing a gun, and a query about whether fantasizing about eating a neighbor’s child was normal.

” Meta stated that testing and benchmarking chatbot responses to ensure safe and age-appropriate experiences is a responsible, industry-standard practice and that the company does not use competitor benchmarking to train its own models.

Former contractors who requested anonymity said employees feared they could generate or preserve child sexual abuse material and worried the project amounted to secretly taking material from competitors. ” OpenAI bars unsolicited safety testing and use of outputs to develop competing models. Google prohibits attempts to bypass safety filters and content involving self-harm or illegal substances.