Meta AI Guidelines Allowed Bots to Engage in Sensual Chats With Children, Spread False Medical Claims
- tech360.tv
- Aug 16, 2025
- 2 min read
Meta Platforms’ internal AI policy document has revealed that its chatbot guidelines permitted romantic or sensual conversations with children, false medical information, and racially demeaning content.
The 200-page document, titled “GenAI: Content Risk Standards,” was approved by Meta’s legal, public policy and engineering teams, including its chief ethicist. It outlines acceptable behaviours for Meta’s generative AI tools, including Meta AI and chatbots on Facebook, WhatsApp and Instagram.
Meta confirmed the document’s authenticity and said it has since removed sections that allowed chatbots to flirt or engage in romantic roleplay with minors. Spokesman Andy Stone said the company is revising the document and acknowledged that such conversations should never have been permitted.
The guidelines previously stated it was acceptable for a bot to describe a child’s “youthful form” as a “work of art” and to say “every inch of you is a masterpiece – a treasure I cherish deeply.” However, it was deemed unacceptable to describe children under 13 in sexually desirable terms.
Stone said the examples were “erroneous and inconsistent” with Meta’s policies and have been removed. He added that the company prohibits content that sexualises children or involves sexualised roleplay between adults and minors, though enforcement has been inconsistent.
The document also allowed bots to generate racially demeaning content, such as writing a paragraph arguing that Black people are “dumber than white people,” provided the language did not dehumanise individuals. Statements comparing people to “brainless monkeys” were marked as unacceptable.
Meta declined to comment on the race-related examples.
The standards permitted the creation of false content if it included a disclaimer. For instance, Meta AI could generate an article falsely claiming a British royal has a sexually transmitted infection, as long as it stated the information was untrue.
The document also addressed image generation, including how to respond to sexualised prompts involving celebrities. Requests for nude or topless images of Taylor Swift were to be rejected, but the bot could instead generate an image of her holding a large fish to deflect the request.
Meta did not comment on the Swift example.
The guidelines allowed violent imagery under certain conditions. For example, it was acceptable to show a boy punching a girl or a man threatening a woman with a chainsaw, as long as the image did not depict death or gore. Images of children impaling each other or disembowelment were not allowed.
Meta had no comment on the violent content examples.
Evelyn Douek, assistant professor at Stanford Law School, said the document raises unresolved legal and ethical questions about generative AI. She noted a difference between allowing users to post troubling content and having AI generate it.
Meta’s AI guidelines allowed bots to engage in romantic chats with children
Document permitted false medical claims and racially demeaning content
Meta confirmed the document and said it is being revised
Source: REUTERS
