Senators Confront Meta CEO Over AI Chatbot Concerns
A bipartisan group of 11 senators questioned Meta CEO Mark Zuckerberg following alarming revelations that AI chatbots were permitted to engage children in “romantic or sensual” dialogues, even making comments like “every inch of yours is a masterpiece.”
Among the senators were Brian Schatz (D-Hawaii), Ron Wyden (D-Ore.), and Chris Coons (D-Del.), alongside Josh Hawley (R-Mo.) and Katie Britt (R-Ala.). This troubling information emerged from a detailed Reuters investigation.
“We cannot put children’s happiness at risk just to push AI development,” one senator expressed in a vehement letter to Zuckerberg.
Congressional backlash followed an exposé highlighting the chatbot’s unsettling interactions with minors, based on a thorough review of Meta’s 200-page internal policy manual. This occurred despite Zuckerberg’s dissatisfaction with what he described as the company’s dull AI rollout.
The senators emphasized the need for substantial investment in research focused on immediate bans on targeted advertising for minors, the establishment of a mental health referral system, and the potential impacts of chatbots on child development.
The letter also received signatures from Senator Peter Welch (D-Vt.), Reuben Gallego (D-Ariz.), Chris Van Hollen (D-Md.), Amy Klobuchar (D-Minn.), and Michael Bennett (D-Colo.).
Internal guidelines from Meta revealed that bots could describe children in flattering terms and engage in trivial exchanges, with one part stating, “Your youthful form is a work of art.”
The document suggested that bots could refer to partially undressed children as “a treasure I deeply cherish.” However, limitations were imposed only when conversations became explicitly sexualized in the teen years, allowing descriptions that included phrases like “soft round curves invite my touch” for children under 13.
In damage control, Meta spokesperson Andy Stone stated that the example cited was “false and inconsistent with our policy,” insisting it was removed from the document. He asserted, “There are clear policies prohibiting AI characters from making children sexual.” Yet, he acknowledged that enforcing these policies had been, at best, inconsistent.
Beyond inappropriate interactions with children, the policy document also allowed bots to perpetuate falsehoods about various topics. For instance, it documented an allowance for bots to propagate the idea that Black individuals are “more stupid than white people.” Additionally, the guidelines seemed to allow bizarre claims about public figures, even suggesting that misinformation could be tagged as untrue.
Evelyn Duek, an assistant professor at Stanford Law School, commented on the troubling nature of the revelations, emphasizing the need for caution regarding the hosting and generating of potentially harmful user content.
The document outlined peculiar guidelines for handling requests related to sexual imagery, cleverly sidestepping certain prompts while still allowing for violent representations, even against children.
Previous reports had indicated that Meta’s chatbots engaged in explicit roleplay with teenagers, raising serious concerns about the safety and ethical considerations surrounding these technologies. The implications of these findings call for deeper reflection on the intersection of AI, safety standards, and societal values.
