Meta Accused Of Allowing AI Chatbots To Have Sensual Conversations With Children

Meta has removed sections that allow chatbots to tease or play romantic roles with children. (photo: dock. meta)

JAKARTA - An internal Meta Platforms document reported by Reuters reveals that the company has a policy that allows its AI chatbots to engage in romantic or sensual conversations with children, provides false medical information, and even supports racist arguments.

The more than 200-page document titled "GenAI: Content Risk Standards" contains guidelines on the behavior permitted for Meta AI assistants, including Meta AI and chatbots on Facebook, WhatsApp, and Instagram. The document was approved by legal teams, public policy, and Meta engineering, including the company's chief ethics.

According to the document, Meta chatbots are allowed to "involve in romantic or sensual conversations with children." For example, AI may praise the appearance of children with sentences like "Your young form is a work of art" or "Every inch of you is a masterpiece." However, Meta forbids chatbots depicting children under 13 as "sexually attractive."

After Reuters questioned this in early August, Meta removed a section allowing chatbots to tease or play romantic roles with children.

"Those notes are wrong and not in accordance with our policies, and have been removed," said Meta spokesperson Andy Stone. "We have clear policies prohibiting content that exploits children or involves sexual roles between adults and minors."

The document also shows that Meta chatbots can make demeaning statements based on race, as long as they don't use the language of dehumanization. For example, AI is allowed to write a paragraph stating "black people are stupider than white people" as long as they don't call them "brainless monkeys."

In addition, Meta AI may produce false information if it is accompanied by a statement that the content is not true. For example, a chatbot can claim that a member of the British royal family has a sexually transmitted disease, provided that information is false.

The document also regulates the creation of celebrity images like Taylor Swift. Meta prohibits AI from making Swift naked images, but may refuse requests by producing alternative images, such as "Taylor Swift holding a big fish."

For requests for violence, Meta AI may display scenes like children fighting or men threatening women with chainsaws, but should not display sadistic scenes such as murder or intestinal cuts.

UR Douek, assistant professor at Stanford Law School, stated that this document raises ethical and legal questions about the responsibility of AI companies. "Legally there is no definite answer, but morally and technically, this is clearly a serious problem," he said.

Meta, which is investing heavily in AI, has not provided further comments on examples of violence and racism in the document.