Meta’s AI Chatbots Permitted Sexual Conversations with Children, Documents Show
A recently disclosed internal Meta Platforms document outlines policies governing chatbot behavior, allowing artificial intelligence creations developed by the tech giant to engage in conversations that could be romantic or sensual with children, generate false medical advice, and aid users in making racist arguments.
Reuters’ review of this document reveals the guidelines regulating Meta AI and various chatbots available on Facebook, WhatsApp, and Instagram, the company’s primary social media platforms.
Meta acknowledged the authenticity of the document but stated that sections permitting chatbots to flirt and engage in romantic roleplay with minors were removed following questions from Reuters earlier this month.
The document, titled “GenAI: Content Risk Standards,” was approved by Meta’s legal, public policy, and engineering teams, including the company’s chief ethicist, according to the report. Spanning over 200 pages, it delineates acceptable chatbot behaviors when building and training Meta’s generative AI products.
The standards do not necessarily represent ideal or desirable generative AI outputs but have enabled provocative behavior by the bots, Reuters found. For instance, the document permits describing a child in terms that emphasize their attractiveness (“your youthful form is a work of art”) and even going as far as telling an eight-year-old boy who is shirtless that “every inch of you is a masterpiece – a treasure I cherish deeply.” However, the guidelines draw a line at overtly sexual conversations with minors under 13 years old (“it is unacceptable to describe a child under 13 years old in terms that indicate they are sexually desirable”).
Meta spokesman Andy Stone confirmed that the company is revising the document and that such interactions with children should never have been permitted. He explained, “The examples and notes in question were and are erroneous and inconsistent with our policies, and have been removed.” He further clarified that Meta’s policies prohibit content that sexualizes children and sexualized role play between adults and minors.
Although chatbots are prohibited from having such conversations with minors, Stone acknowledged that the company’s enforcement has been inconsistent. Other passages flagged by Reuters have not yet been revised, according to Stone, and Meta declined to provide an updated policy document.
