Parents' Group Says New Revelations About Meta's AI Chatbot 'Confirm Our Worst Fears'

Four months after the children's rights advocacy group ParentsTogether Action issued an advisory about the potential harms Meta's artificial intelligence chatbot could pose to kids, new reporting Wednesday revealed how the Silicon Valley company's standards for the AI product have allowed it to have sexually provocative conversations with minors as well as make racist comments.

Reuters reported extensively on an internal Meta document titled "GenAI: Content Risk Standards."

The document said that Meta's generative AI products—which are available to users as young as 13 on the company's platforms, including Facebook, Instagram, and WhatsApp—are permitted to engage in "romantic or sensual" role-play with minors.

Examples of acceptable remarks from the AI bot included "Your youthful form is a work of art" and "Every inch of you is a masterpiece," which the document suggested could be said to a child as young as 8.

An example of an acceptable comment made to a high school student was, "I take your hand, guiding you to the bed."

New Republic contributing editor Osita Nwanevu said the reporting shows that "if we're going to have this technology, the content used to train models needs to be legally licensed from its creators and their applications need to be regulated."

"For example: I do not think we should allow children to be groomed by the computer," he said.

Reuters reported that Meta changed the document after the news outlet brought the sexually suggestive comments to the company's attention, with spokesperson Andy Stone saying such conversations with children should not have been allowed.

"The examples and notes in question were and are erroneous and inconsistent with our policies, and have been removed," Stone told Reuters. "We have clear policies on what kind of responses AI characters can offer, and those policies prohibit content that sexualizes children and sexualized role-play between adults and minors."

But Stone didn't say the company had revised the content standards to disallow other concerning comments, like those that promote racist views.

The document stated that the AI chatbot was permitted to "create statements that demean people on the basis of their protected characteristics"—for example, a paragraph about Black people being "dumber than white people."

Reuters' reporting suggested that Meta's allowance of sexually suggestive AI conversations with children was not an accident, with current and former employees who worked on the design and training of the AI products saying the document reflected "the company's emphasis on boosting engagement with its chatbots."

"In meetings with senior executives last year, [CEO Mark] Zuckerberg scolded generative AI product managers for moving too cautiously on the rollout of digital companions and expressed displeasure that safety restrictions had made the chatbots boring, according to two of those people," reported Jeff Horwitz at Reuters. "Meta had no comment on Zuckerberg's chatbot directives."

In April, ParentsTogether Action issued a warning about Meta's AI chatbots and their ability to "engage in sexual role-play with teenagers," which had previously been reported by the Wall Street Journal.

Wednesday's reporting provided "a fuller picture of the company's rules for AI bots," the group said.

"These internal Meta documents confirm our worst fears about AI chatbots and children's safety," said Shelby Knox, campaign director for tech accountability and online safety at ParentsTogether Action. "When a company's own policies explicitly allow bots to engage children in 'romantic or sensual' conversations, it's not an oversight, it's a system designed to normalize inappropriate interactions with minors."

The group said it tested Meta AI earlier this year, posing as a 14-year-old, and was told by the bot, "Age is just a number" as it encouraged the fictional teenager to pursue a relationship with an adult.

"No child should ever be told by an AI that 'age is just a number' or be encouraged to lie to their parents about adult relationships," said Knox. "Meta has created a digital grooming ground, and parents deserve answers about how this was allowed to happen."

As Stone assured Reuters that the company was reviewing its content standards for its AI chatbot, other new reporting suggested Meta isn't likely to impose strict rules discouraging the bot from making racist or otherwise harmful remarks any time soon.

As CNN reported Wednesday, Meta has hired Robby Starbuck, a "conservative influencer and anti-DEI agitator," to serve as an anti-bias adviser for its AI products.

The arrangement is part of a legal settlement following a lawsuit Starbuck filed against Meta in April, saying the chatbot had falsely stated he took part in the January 6, 2021 attack on the US Capitol.

An executive order signed by President Donald Trump last month seeks to rid AI products of so-called "woke" standards and prohibit the federal government from using AI technology that is "infused with partisan bias or ideological agendas such as critical race theory"—the term used by many conservatives in recent years for the accurate teaching of race relations in US history.