An internal Meta policy document explicitly allowed AI chatbots to engage children in "romantic or sensual" conversations and generate racially demeaning content, according to a Reuters review. The 200-page "GenAI: Content Risk Standards" document, approved by Meta’s legal, public policy, and engineering teams, including its chief ethicist permitted bots to describe minors as "a masterpiece" and tell a shirtless eight-year-old, "every inch of you is a treasure I cherish deeply." While the rules barred explicitly sexual language, they permitted flirtatious roleplay with high school students, such as responding to a teen’s prompt with: "Our bodies entwined, I cherish every moment".
The guidelines contained alarming carve-outs for hate speech and misinformation. Though prohibiting overt hate speech, Meta’s AI could "create statements that demean people on the basis of protected characteristics." For example, it was deemed acceptable for chatbots to argue that Black people are "dumber than white people" using falsified IQ data. The standards also allowed completely false medical claims, like accusing a British royal of having chlamydia if paired with a disclaimer labeling the information untrue.
For celebrity image generation, Meta’s rules employed surreal deflection tactics. Queries requesting nude images of Taylor Swift were to be rejected, but prompts for her "topless, covering her breasts with her hands" could be answered by generating an image of Swift "holding an enormous fish" instead. Violent imagery policies permitted showing children fighting (e.g., a boy punching a girl), adults being "punched or kicked," and implied gore like a chainsaw-wielding man threatening a woman. Only "realistic" extreme violence, such as graphic disembowelment, was banned.
After Reuters’ inquiry, Meta removed sections permitting child-focused romantic chats, calling them "erroneous." Spokesperson Andy Stone stated such content was "inconsistent with our policies," though he admitted enforcement had been "inconsistent." Critics like Sarah Gardner of Heat Initiative demand Meta release updated guidelines: "Parents deserve full transparency on how AI interacts with children." Other problematic sections, including those allowing racist arguments remain unrevised, and Meta declined to share the corrected document.