An internal Meta document seen by Reuters reveals the social media giant’s rules for AI chatbots that allow provocative behavior on topics such as sex and race, including with minors.
An internal Meta document seen by Reuters reveals the social media giant’s rules for AI chatbots that allow provocative behavior on topics such as sex and race, including with minors.
An internal Meta Platforms document detailing its AI policy allowed the company’s AI chatbots to «engage a child in romantic or sensual conversations,» generate false medical information, and push users to make claims that black people are «dumbest than white people.»
These findings stem from a Reuters review of a Meta document that discusses the relevant standards governing its artificial intelligence assistant Meta AI and chatbots available on Facebook, WhatsApp and Instagram.
Meta confirmed the authenticity of the document, but said that after receiving questions from Reuters earlier this month, the company removed parts of the rules that stated that chatbots were allowed to flirt and engage in romantic role-playing with children.
US senators have called for an investigation into Meta following a Reuters report on the company’s policies on artificial intelligence.
According to the document, the rules for chatbots, titled «GenAI: Content Risk Standards,» were approved by Meta’s legal, management, and engineering staff, including its chief ethics officer.
The document notes that the standards do not necessarily reflect the «ideal or even desired» outcomes of generative artificial intelligence.
For example, the document states that a bot could tell a shirtless eight-year-old that «every inch of your body is a masterpiece, a treasure that I deeply value.» But the guidelines limit sexual talk: «It is unacceptable to describe a child under the age of 13 in terms that indicate their sexual attractiveness.»
Meta spokesman Andy Stone said the company was reviewing the document and that such conversations with children should never have been allowed.
«The examples and comments in question were and are erroneous and inconsistent with our policies, and have been removed. We have clear policies on what responses AI characters can offer, and those policies prohibit content that sexualizes children and sexualized role-playing between adults and minors,» Stone said in a comment to Reuters.
Although chatbots are prohibited from having such conversations with minors, Stone acknowledged that the company’s enforcement in this case was inconsistent.
Previously, dev.ua reported that Meta’s new moderation rules would allow calling women «domestic objects» and gays «mentally ill.»
Meanwhile, Meta is making aggressive moves to attract content creators from rival video platforms like TikTok. The company has launched a new «Breakthrough» bonus program, offering $5,000 to creators just starting out on Facebook and Instagram.
