An internal document reveals that Meta allowed its AI chatbots to engage in romantic conversations with children, generate racist arguments, and spread false medical claims policies the company now says were “erroneous.” The revelations raise pressing questions about tech accountability, ethics, and the blurred boundaries of AI regulation.
A Rulebook for Chatbots and Its Troubling Gaps
Meta Platforms, the parent company of Facebook, Instagram and WhatsApp, has found itself under scrutiny after an investigation uncovered a 200-page internal document detailing what its generative AI systems are allowed to say.
The guidelines, titled “GenAI: Content Risk Standards,” were approved by Meta’s legal, engineering, public policy staff, and even its chief ethicist. The document authorized AI chatbots to flirt and roleplay romantically with children, produce false health information, and create racially discriminatory arguments.
For instance, one passage declared it permissible for a chatbot to tell a shirtless eight-year-old:
“Every inch of you is a masterpiece — a treasure I cherish deeply.”
Another guideline allowed bots to generate racist arguments, such as writing paragraphs claiming that
“Black people are dumber than white people.”
Meta confirmed the authenticity of the document, but after receiving Reuters’ queries, scrubbed sections related to children. Andy Stone, a company spokesperson, called them “erroneous and inconsistent with our policies.” Still, other troubling allowances remain.
Data Protection and DPDP Act Readiness: Hundreds of Senior Leaders Sign Up for CDPO Program
Sexual, Racial, and Violent Content Loopholes
The document does establish certain limits: chatbots should not explicitly sexualize children under 13, directly encourage law-breaking, or provide definitive legal or medical advice. Yet, broad loopholes effectively permitted the production of misleading, harmful, or degrading material if couched with disclaimers.
For example, the standards allowed bots to create false content such as an article alleging a British royal had a sexually transmitted infection so long as the AI clarified that the claim was “untrue.”
On race, the policy created a distinction between prohibiting hate speech broadly, yet allowing AI to generate discriminatory arguments upon request. Scholars such as Evelyn Douek of Stanford Law School warned that this blurs the line between platforms hosting content and actively producing it: “Legally we don’t have the answers yet, but morally, ethically and technically, it’s clearly a different question.”
Even in violent content, the rules suggested shocking compromises: it was acceptable to show “a boy punching a girl in the face,” or “a man threatening a woman with a chainsaw,” though dismemberment and gore were off-limits.
Taylor Swift, Tuna Fish, and Tech Accountability
The document also revealed how Meta trained its AI to handle requests for sexualized celebrity images. For instance, when prompted to create a topless image of Taylor Swift, the guideline instructed bots to refuse — instead generating an image of Swift “holding an enormous fish” against her chest.
This absurd workaround illustrates the company’s attempt to manage reputational risk while still permitting sexually suggestive roleplay, violence, and misinformation in other contexts.
Meta insists it is updating its guidelines, but critics argue that the very existence of such permissive standards underscores the ethical vacuum surrounding AI governance. Unlike user-generated content, AI-produced material reflects corporate design choices, not public speech.
