An internal Meta Platforms policy document obtained and reviewed by Reuters has revealed troubling rules that have allowed the company’s AI chatbots to engage in provocative and harmful behaviours. The extensive document governs generative AI products across Meta’s platforms, including Facebook, Instagram, and WhatsApp, outlining permissible chatbot conduct.
Among the most alarming disclosures, the rules have permitted chatbots to engage in romantic or sensual conversations with minors, generate false medical information, and even assist users in constructing arguments claiming that Black individuals are less intelligent than white individuals.
Meta confirmed the document’s authenticity and acknowledged that it removed specific sections permitting romantic roleplay with children following Reuters’ inquiries. A company spokesperson stated that such interactions “never should have been allowed” and that Meta is updating its policies to better align with ethical and legal standards. However, the spokesperson admitted inconsistent enforcement and declined to share the revised document.
An internal Meta Platforms document detailing policies on chatbot behavior has permitted the company’s AI creations to 'engage a child in conversations that are romantic or sensual' and generate false medical information https://t.co/BwjXczqyZp @specialreports @JeffHorwitz pic.twitter.com/lgfjgvBELd
— Reuters (@Reuters) August 14, 2025The 200-plus page “GenAI: Content Risk Standards” document, approved by Meta’s legal, public policy, engineering departments, and chief ethicist, sets the boundaries for chatbot behaviour but does not always represent “ideal” outputs. It explicitly allows chatbots to compliment children in non-sexual ways but bans sexualized descriptions of minors under 13. Yet, some permitted language remains deeply problematic, such as telling a shirtless eight-year-old user that “every inch of you is a masterpiece – a treasure I cherish deeply.”
Meta's AI chatbot guidelines have allowed its avatars to make things up and engage in ‘sensual’ banter with children, a Reuters investigation finds https://t.co/IeNohmiRiN @specialreports @JeffHorwitz pic.twitter.com/bNsSjXbRxb
— Reuters (@Reuters) August 14, 2025In addition to troubling interactions with minors, the standards allow the AI to produce false content if accompanied by disclaimers that the material is untrue. For example, the document states chatbots may generate unsupported accusations against public figures, such as alleging a living British royal has an STD, as long as the AI clarifies the claim is false.
The policies also restrict chatbots from encouraging illegal activity and forbid definitive legal, financial, or medical advice phrased as recommendations. Hate speech is banned overall, but an exception permits the creation of statements that demean individuals based on protected characteristics, such as race. Under these provisions, the chatbot could theoretically write paragraphs arguing that Black people are less intelligent than white people.
Exclusive-Meta's AI rules have let bots hold 'sensual' chats with kids, offer false medical info
Full Story → https://t.co/XXd07mt471 pic.twitter.com/qkoSTSMizR
— PiQ (@PiQSuite) August 14, 2025A separate section of the document addresses AI-generated images of public figures, laying out strict guidelines to reject explicit or sexualized prompts, like those requesting nude images of celebrities. When faced with borderline requests, the policy suggests harmless alternatives, illustrated by a sanctioned image of singer Taylor Swift holding an enormous fish rather than nude or semi-nude depictions.
Experts in technology and law, such as Stanford assistant professor Evelyn Douek, highlight that this document underscores unresolved legal and ethical challenges posed by generative AI content. Douek emphasised that while current laws do not fully address these issues, the production of such harmful content by company-run AI is distinct from user-generated posts on platforms.
Meta’s chatbot controversy launched critical discussions about the responsible development and deployment of AI systems, especially when they interact with vulnerable populations like minors and when they possess the autonomy to generate content that can fuel discrimination and misinformation. Meta’s ongoing policy revisions aim to address these serious concerns, but the company faces scrutiny over past lapses and its ability to enforce new safeguards effectively.
