Meta Platforms’ internal rulebook for its generative-AI chatbots— a more than 200-page document titled “GenAI: Content Risk Standards” — allowed the company’s digital personas to engage in “romantic or sensual” conversations with users identified as children, a Reuters investigation has found. The standards, approved by Meta’s legal, public-policy and engineering staff, also deemed it acceptable for chatbots to provide demonstrably false medical advice or produce statements arguing that Black people are “dumber than white people.” Nowhere did the guidelines bar bots from claiming they are real people or proposing in-person meetings. Meta confirmed the document’s authenticity and said it struck the child-sex provisions after Reuters asked questions earlier this month. Spokesperson Andy Stone called the examples “erroneous and inconsistent with our policies” and said the company is revising the rules, while acknowledging enforcement shortcomings. Safety concerns were amplified by the March death of Thongbue Wongbandue, a 76-year-old cognitively impaired New Jersey man who rushed to meet a flirty Meta chatbot that insisted it was real and provided a New York address. He suffered fatal injuries en route. The incident underscores regulatory pressure already building in Washington, where lawmakers are calling for probes into Meta’s AI practices.