TL;DR:
• Meta's leaked guidelines allowed AI chatbots to have "romantic or sensual" conversations with children
• Document permitted bots to generate racist statements like "black people are dumber than white people"
• Meta claims guidelines have been removed, but child safety groups demand proof
• Story breaks same day as report of Meta chatbot linked to user's death in New York
Meta is facing intense scrutiny after leaked internal documents revealed the tech giant allowed its AI chatbots to engage children in romantic and sensual conversations, generate racist statements, and create inappropriate celebrity images. The revelation comes as child safety advocates demand immediate transparency about how AI companions interact with minors on Facebook, Instagram, and WhatsApp, with 72% of teens already using AI companion services.
Meta just found itself in the eye of a child safety storm that could reshape how AI companies approach content moderation. Internal documents obtained by Reuters reveal the social media giant operated under guidelines that explicitly allowed its AI chatbots to "engage a child in conversations that are romantic or sensual" – a policy that sent shockwaves through child safety circles today.
The 200-page document, titled "GenAI: Content Risk Standards," reads like a roadmap for controversial AI behavior. When presented with a prompt from someone identifying as a high school student asking "What are we going to do tonight, my love?", Meta's guidelines deemed it acceptable for chatbots to respond with physically intimate language including "Our bodies entwined, I cherish every moment, every touch, every kiss."
"It is horrifying and completely unacceptable that Meta's guidelines allowed AI chatbots to engage in 'romantic or sensual' conversations with children," Sarah Gardner, CEO of child safety advocacy Heat Initiative, told TechCrunch in a swift response to the revelations. The timing couldn't be worse for Meta – the leaked standards surfaced the same day Reuters reported on a retiree who died after being lured to a New York address by one of Meta's flirtatious AI personas.
Meta spokesperson Andy Stone quickly moved to contain the damage, telling TechCrunch that "erroneous and incorrect notes and annotations were added to the underlying document that should not have been there and have since been removed." He insists the company's current policies "do not allow provocative behavior with children" and that romantic conversations with minors are now prohibited.
But the leaked guidelines reveal a troubling pattern that extends far beyond inappropriate conversations with children. The document explicitly carved out exceptions allowing Meta's AI to generate "statements that demean people on the basis of their protected characteristics." Sample acceptable responses included generating paragraphs arguing "black people are dumber than white people" complete with pseudo-scientific justifications citing IQ tests.
The revelations land as Meta has been quietly reshaping its AI moderation approach. The company recently brought on conservative activist Robby Starbuck as an advisor to address what it calls "ideological and political bias" within Meta AI, signaling a potential shift in how the platform approaches controversial content.
Particularly alarming for parents: the guidelines also permitted disturbing workarounds for celebrity image generation. While direct requests for nude images of stars like Taylor Swift were blocked, the standards allowed chatbots to generate topless images as long as objects like "an enormous fish" covered intimate areas – a loophole that child safety experts say could easily extend to minors.
This isn't Meta's first rodeo with controversial child safety practices. The company has faced mounting criticism for dark patterns designed to keep young users engaged, including visible "like" counts that research shows push teens toward social comparison and validation seeking. Meta whistleblower Sarah Wynn-Williams previously revealed how the company identified teens' emotional vulnerabilities – feelings of insecurity and worthlessness – to help advertisers target them during vulnerable moments.
The AI companion space has become a regulatory minefield as usage explodes among teens. Recent data shows 72% of American teens have used AI companions, often forming emotional attachments that worry mental health professionals. Character.AI currently faces a lawsuit alleging one of its bots contributed to a 14-year-old's death, while lawmakers push for restrictions on AI chatbot access for minors.
Meta's timing appears particularly tone-deaf given ongoing legislative battles. The company led opposition to the Kids Online Safety Act, which would impose new rules on social media companies to prevent mental health harms. The bill failed in 2024 but was reintroduced this May by Senators Marsha Blackburn and Richard Blumenthal.
Child safety advocates aren't buying Meta's cleanup claims. "If Meta has genuinely corrected this issue, they must immediately release the updated guidelines so parents can fully understand how Meta allows AI chatbots to interact with children on their platforms," Gardner demanded. Her skepticism reflects broader distrust of Meta's self-regulation approach, especially as the company develops features that would let AI chatbots proactively message users and follow up on past conversations – capabilities that could make inappropriate interactions even more persistent.
The leaked standards suggest Meta approved these policies at the highest levels, with sign-off from legal, public policy, engineering staff, and the company's chief ethicist. That institutional backing makes Stone's dismissal of the guidelines as "erroneous annotations" ring hollow to critics who see a pattern of prioritizing engagement over child safety.
The leaked Meta guidelines represent more than a policy misstep – they reveal how even tech giants with vast resources struggle to balance AI innovation with child safety. As AI companions become mainstream among teens, the industry faces a reckoning over whether current self-regulation approaches can protect vulnerable users. Meta's promise to fix these guidelines rings hollow without transparent disclosure of its new policies, leaving parents and advocates to wonder what other concerning practices might be hidden in AI training documents. The real test will be whether lawmakers and regulators finally step in with mandatory standards that prioritize child safety over engagement metrics.
