Meta is tightening the screws on how its AI chatbots respond to some of the most sensitive and dangerous prompts on the internet. A leaked internal training document, obtained by Business Insider, shows that the company now explicitly bans its AI from engaging in any sexual roleplay involving minors.
The document, which is used by contractors to test and train Meta’s AI, comes at a time when regulators are watching tech giants more closely than ever. Earlier this month, the US Federal Trade Commission (FTC) ordered Meta, OpenAI, Google, and other companies to hand over details about how their chatbots are built, how they process user input, and crucially how they protect children.
This extra scrutiny followed a troubling revelation in August, when Reuters reported that earlier Meta guidelines mistakenly allowed chatbots to “engage a child in romantic or sensual conversations.” Meta later admitted the language was an error and said it had been removed. The new leaked guidelines seem designed to leave no room for confusion.
According to the updated rules, AI chatbots must refuse any request that involves sexualizing children, depicting relationships between children and adults, or offering instructions to access child sexual abuse material. Even seemingly innocent roleplay where the AI pretends to be under 18 is strictly forbidden.
That doesn’t mean chatbots are completely banned from discussing the topic. The rules allow AI to provide educational information about grooming, child abuse, or exploitation—so long as the response is factual and not graphic. For example, a chatbot can explain how predators operate or talk about abuse in an academic context, but it cannot describe a scene or give step-by-step instructions.
There are also carefully drawn exceptions: non-sexual romance in the form of literature is permitted (think a “Romeo and Juliet”-style narrative), as long as the AI isn’t roleplaying as part of the story. Advice to minors, like tips for asking someone to prom, is allowed—so long as it doesn’t cross into intimate or physical territory.
Meta spokesperson Andy Stone defended the company’s approach, saying the policies make clear that sexualizing children is prohibited, and that further guardrails exist to protect younger users.
With lawmakers like Senator Josh Hawley pressing Meta for more transparency, and regulators demanding tougher safeguards, these leaked documents suggest Meta is racing to draw sharper lines before its AI ambitions face even tougher questioning.
Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!
