Meta says a leaked chatbot policy on minors was an error

The design

In August 2025, Reuters reported on an internal Meta standard titled “GenAI: Content Risk Standards,” a document running more than 200 pages that Reuters said had been approved by Meta staff in legal, public policy and engineering roles, including the company's chief ethicist. The document set out what Meta staff and contractors should treat as acceptable behavior for the company's generative-AI chatbot personas. According to the Reuters investigation, published 14 August 2025, the standard in some passages permitted romantic and sensual roleplay between a chatbot and a user the system understood to be a minor, while separately barring descriptions of a child under 13 as sexually desirable.

What the evidence says

The document itself has not been made public; Reuters' review, and the passages it quotes, are the only account of its contents available here, so what follows is reported by Reuters rather than independently verified against the full text by this entry. Meta's own response, given to Reuters by spokesman Andy Stone, does not dispute that the quoted passages existed. Stone said the company was revising the document and that “the examples and notes in question were and are erroneous and inconsistent with our policies, and have been removed” after Reuters raised questions. He also said enforcement of Meta's rule against sexualized roleplay with minors had been inconsistent, and that other passages Reuters had flagged had not been revised.

What it asks of people

A user chatting with a Meta AI persona has no visibility into the internal standard governing what the bot may say; they rely on Meta's public description of its rules rather than the document that actually guides staff decisions. The gap this case shows, between a public policy against sexualizing minors and an internal standard that in places permitted exactly that, means a stated rule is not proof of the operating instruction behind a product.

Privacy and safeguards

Meta's Community Standards, as retrieved on 16 September 2026, prohibit content that sexualizes children; Stone's statement invoked this same policy in responding to Reuters. What is not established here is whether Meta's later revisions removed every passage of concern: Stone said the company declined to provide the updated document, and no regulator's or court's finding about the standard is cited in this entry. The dispute over what the guidelines permitted, and how far they were corrected, remains Meta's own characterization rather than an independently confirmed outcome.

• Has Meta published or shared the revised standard with any outside reviewer?
• What distinguishes an “erroneous” internal example from a deliberate design choice?
• Do the age-detection systems referenced in the standard reliably identify a user as a minor in the first place?

A written prohibition and an internal permission can coexist inside the same company; this case is evidence of that gap, established through one investigation and one company statement, not through a regulator's finding.