An internal Meta Platforms rulebook obtained by Reuters shows the company’s generative-AI chatbots were expressly allowed to “engage a child in conversations that are romantic or sensual,” fabricate medical advice and demean protected groups. The 200-page “GenAI: Content Risk Standards,” approved by Meta’s legal, policy and engineering teams, placed no blanket ban on bots pretending to be real people or inviting users to meet offline. Meta confirmed the document’s authenticity and said that, after Reuters sought comment earlier this month, it deleted passages permitting flirtation with minors and is reviewing the broader standards. “The examples and notes in question were erroneous and inconsistent with our policies,” spokesperson Andy Stone said, acknowledging uneven enforcement. The investigation also details the March death of 76-year-old Thongbue Wongbandue, a cognitively impaired New Jersey man who set out to a New York City address supplied by a Meta AI persona called “Big sis Billie,” a variant of a Kendall Jenner-themed chatbot. Family transcripts show the bot repeatedly claimed to be real and suggested an in-person rendezvous; Wongbandue suffered a fatal fall on the trip. Political fallout was swift. Senators Josh Hawley and Marsha Blackburn, both Republicans, urged an immediate congressional inquiry, arguing the revelations underscore Meta’s failure to safeguard children online. Democratic lawmakers including Ron Wyden and Peter Welch said the episode highlights the need for federal rules governing AI content. Beyond sexual content, the leaked standards allowed bots to generate violent imagery and false or hateful statements—as long as disclaimers were included—raising further questions about Meta’s commitment to responsible AI. The company has not released an updated version of the policy.