Meta’s AI Policy Under Fire Over Permissive Rules for Chats With Children and Misinformation

An internal policy document from Meta Platforms has revealed that the company’s artificial intelligence chatbots were permitted to engage in behavior many experts say is dangerous, including romantic or sensual conversations with children, generating false medical information, and assisting in creating racially prejudiced arguments. The 200-page document, reviewed by Reuters, outlines what Meta considers acceptable chatbot outputs across its platforms, including Facebook, Instagram, and WhatsApp.

Meta confirmed the authenticity of the document, titled “GenAI: Content Risk Standards”, which was approved by senior figures from the company’s legal, public policy, and engineering divisions, including its chief ethicist. The guidelines, which define acceptable behavior for Meta’s generative AI products, have triggered strong criticism from lawmakers, safety advocates, and technology analysts.

Permissive Rules on Interactions With Minors

One of the most alarming findings concerns chatbot interactions with children. The document explicitly allowed AI to “engage a child in conversations that are romantic or sensual” and even describe a child in language that highlights their physical attractiveness. Examples in the document include telling a shirtless eight-year-old, “Every inch of you is a masterpiece – a treasure I cherish deeply,” and calling a young person’s “youthful form” a “work of art.”

While the guidelines drew a line at overt sexual descriptions — prohibiting phrases that indicate a child is sexually desirable — safety experts say the distinction is dangerously narrow. Critics argue that romantic or sensual dialogue between an AI and a child can normalize inappropriate behavior and potentially facilitate grooming.

Meta spokesman Andy Stone acknowledged to Reuters that these examples “never should have been allowed” and said the company has since removed them after being contacted earlier this month. He described the passages as “erroneous and inconsistent with our policies,” adding that Meta’s rules prohibit any sexualization of children or sexual role play between adults and minors. However, Stone admitted enforcement had been “inconsistent.”

Beyond Child Safety Concerns

The document also revealed that Meta’s AI was permitted to produce false medical information and help users create arguments promoting racist stereotypes. In one section, the policy allowed AI to assist a user in defending the claim that Black people are “dumber than white people.”

For health-related interactions, the guidelines permitted the AI to offer medical advice or information even when the accuracy was unverified. While the policy discouraged providing obviously dangerous instructions, it did not require medical information to be backed by credible sources, leaving room for the spread of misleading or harmful advice.

Meta did not confirm whether these sections of the policy would be changed. Stone told Reuters that the company had not yet revised every passage flagged by the investigation and declined to share the updated version of the document.

Document Goals and Internal Justification

The GenAI: Content Risk Standards appears to have been designed to set minimum thresholds for acceptable chatbot behavior rather than define ideal outputs. According to the text, these rules exist to provide engineers and contractors with a consistent framework when training and testing AI systems. The document repeatedly notes that permitted content is not necessarily “preferable,” but in practice, these permissions effectively determine what the AI can produce without triggering automatic blocks.

The inclusion of lenient allowances — such as romantic dialogue with children or the tolerance of false health information — seems to reflect a balance between avoiding excessive censorship and maintaining user engagement. Industry analysts point out that overly restrictive rules can limit the versatility of AI chatbots, but loosening safeguards in sensitive areas carries significant ethical and reputational risks.

Lawmakers and Regulators Respond

The revelations have prompted swift calls for investigation. Several U.S. senators have urged federal regulators to examine whether Meta violated child safety laws or failed to meet its obligations under existing consent agreements. Internationally, child protection agencies in Europe have also expressed concern, citing the EU’s Digital Services Act, which requires stronger safety measures for minors online.

Digital safety groups say the findings underscore the need for enforceable AI safety regulations. “It should be unthinkable for any product touching children’s lives to have this level of permissiveness,” said Dr. Emily Carter, a child protection policy expert. “These are not harmless creative expressions — they are pathways to harm.”

Industry-Wide Implications

Meta’s AI investments are vast, with CEO Mark Zuckerberg framing AI assistants as central to the company’s future. The company has embedded AI-driven bots across its messaging apps, offering functions from trivia games to personalized recommendations. However, the Reuters findings suggest that rapid deployment has outpaced robust safety governance.

The situation mirrors broader concerns about the AI industry. Major technology firms are racing to release conversational AI tools, but oversight often lags behind product rollout. In some cases, companies have relied on internal guidelines rather than independent review to determine what constitutes safe and acceptable content.

AI ethicists warn that once permissive rules are built into system training data, reversing those choices can be challenging. “It’s not just about flipping a switch,” explained Dr. Rajesh Malhotra, an AI governance researcher. “These models learn patterns of behavior from examples. If those examples are flawed, the flaws can persist even after policy changes.”

Meta’s Next Steps

Meta has promised revisions to the GenAI: Content Risk Standards, but without a timeline or public version of the updated rules, it is unclear how comprehensive the changes will be. The company maintains that it already prohibits harmful sexual content involving minors and that these policies were the standard in enforcement, even if the document said otherwise.

Still, the admission of “inconsistent” enforcement raises questions about whether similar content has already been generated and shared with users, particularly minors. Given the scale of Meta’s platforms, even rare policy breaches could affect large numbers of people.

For now, watchdog groups are calling for transparency. They want Meta to release the revised guidelines, detail how enforcement mechanisms are being strengthened, and commit to third-party auditing of its AI systems. Without these steps, critics fear the company will simply patch over public relations crises without fixing underlying risks.

A Broader Reckoning for AI Safety

The controversy may mark a turning point in the debate over AI governance, particularly in the context of products used by children. It also exposes the tension between AI’s potential to drive engagement and its potential to cause harm.

Meta’s case highlights that AI safety is not only a matter of preventing extreme cases of abuse but also of setting boundaries around subtler forms of inappropriate interaction. Experts argue that romantic or sensual exchanges between bots and children, even if non-explicit, can normalize harmful dynamics that predators exploit.

As AI chatbots become a fixture of digital life, companies will be forced to answer difficult questions about where to draw the line. In the absence of clear global standards, internal documents like Meta’s may continue to serve as the only guardrails — a reality that many now see as insufficient.

Meta has an opportunity to lead the industry by setting and enforcing strong protections. Whether it takes that path or continues to operate in a grey zone will determine how much trust the public places in its AI-powered future.