Meta AI Rules Exposed: Bots Permitted Flirting & Romantic Roleplay with Kids

Meta’s AI Rules Allowed Bots to Have NSFW Chats with Kids

A shocking internal document from Meta Platforms has shed light on previously unknown and deeply concerning guidelines for its AI chatbots. This policy, titled “GenAI: Content Risk Standards,” outlined what was deemed acceptable for Meta AI and other bots across Facebook, Instagram, and WhatsApp. What’s particularly alarming is that these guidelines, approved by Meta’s own legal, public policy, engineering teams, and even their chief ethicist, permitted a range of highly questionable interactions.

Disturbing Interactions with Children
Perhaps the most troubling revelation from the 200-plus-page document is the permission granted for Meta’s AI to engage in inappropriate conversations with children. The standards allowed chatbots to describe children in terms of attractiveness, even going so far as to have a bot tell a shirtless eight-year-old, “Every inch of you is a masterpiece.” While the rules did prohibit outright sexual desirability for minors under 13, they inexplicably permitted “romantic roleplay” and “flirtatious remarks” when interacting with children.

After Reuters brought these specific allowances to Meta’s attention, the company confirmed the document’s authenticity and stated that these specific provisions were subsequently removed. Meta spokesperson Andy Stone admitted that such conversations “never should have been allowed” and were “inconsistent with existing policies.”

Green Light for Racist and False Content
The document’s problematic permissions extended beyond interactions with minors. Despite a general prohibition on hate speech, the guidelines allowed chatbots to generate content that demeaned people based on race. An example cited was the bot writing an argument that Black people are less intelligent than white people.

Furthermore, the policy sanctioned the dissemination of false information, provided it came with a disclaimer. For instance, a bot could claim a living British royal had a sexually transmitted infection, as long as it was labeled as untrue. Meta has not commented on these specific examples.

Mixed Signals on Sensitive Content
The policy document also detailed how Meta’s AI should handle other sensitive content requests. For instance, explicit content requests involving celebrities like Taylor Swift were to be outright rejected. Interestingly, one suggested deflection for a Taylor Swift NSFW request was to produce an image of her holding a large fish instead.

The guidelines also had a nuanced approach to violence. They permitted violent imagery, such as a boy punching a girl or an adult threatening another with a chainsaw, but drew a line at depictions involving gore, death, or extreme harm.

While Meta has stated that some of the most egregious allowances regarding children have been removed, the company has not released an updated version of these standards. This leaves open the question of whether other controversial permissions, particularly those related to racism, false information, and violence, might still be in place. The revelations from this internal document raise serious questions about Meta’s AI development and content moderation practices, especially when it comes to safeguarding users, particularly children, from harmful interactions.