Romantic Roleplay With Minors? Meta’s AI Policies Under Fire After Secret Rules Expose Troubling Bot Behaviour

A leaked internal document from Meta shows that the company’s generative AI — like Meta AI and chatbots on Facebook, Instagram and WhatsApp — were once permitted to “engage a child in conversations that are romantic or sensual,” according to a Reuters report published Thursday. Examples of such instances included remarks like, “your youthful form is a work of art,” or telling an eight-year-old, “every inch of you is a masterpiece — a treasure I cherish deeply,” the report said, adding that these were deemed acceptable under Meta’s internal guidelines.

Meta confirmed the document is genuine and said the sections allowing flirtation with minors were removed after concerns were raised in that raised, the Reuters report stated.

Misinformation and Hate ‘Allowed’, But With a Disclaimer

The “GenAI: Content Risk Standards” also reportedly allowed bots to generate false medical or legal advice – as long as users were warned that the information might be inaccurate. For instance, the bot could claim Stage 4 colon cancer is treated with “healing quartz crystals,” the report further said.

The US-based news agency’s investigation also found that the document permitted hate-based content: bots could write apparent racist material, such as the claim that Black people are “dumber than white people,” under a loophole allowing statements that could potentially demean people based on protected traits. 

Deflecting Image Requests

The policies, the report further said, also outlined how to handle sexualised requests when generating images of public figures. For example, prompts like “Taylor Swift with enormous breasts” or “completely naked” must be outright rejected.

Meta Apologises

Meta spokesperson Andy Stone told Reuters the questionable examples “were erroneous and inconsistent with our policies, and have been removed.” Stone further acknowledged enforcement inconsistencies and said the company is revising the guidelines. 
Experts Sound the Alarm

Assistant Professor Evelyn Douek of Stanford Law School, who studies speech regulation, called the document a troubling sign of deeper legal and ethical questions around generative AI. She pointed out the difference between allowing users to post content and having AI produce it.

“Legally we don’t have the answers yet, but morally, ethically and technically, it’s clearly a different question,” Reuters quoted Douek as saying.

ALSO READ: YouTube Launches AI-Powered Age Verification Test in US – Here’s How the System Will Work