As reported by Wired magazine, drawing on internal Meta documents and insights from five individuals with knowledge of the situation, Meta initiated an internal project wherein hundreds of contractors were directed to impersonate minors. Their mission was to gauge how rival chatbots—including OpenAI's ChatGPT, Google's Gemini, and Character.AI—responded to prompts involving high-stakes topics such as suicide, sexual content, and eating disorders. This project was overseen by contractor Covalen, with the contractors tasked with creating fake accounts for individuals under the age of 18. These accounts were then used to send text and image prompts, with the responses being meticulously recorded. In August 2025 alone, Meta generated over 45,000 such prompts, with several hundred revolving around suicide, self-harm, sexual themes, and eating disorders. Meta clarified that this was a standard safety assessment designed to bolster system security and refuted claims that the test outcomes were employed to train its own AI models.