Meta docs show its AI chatbots were allowed to flirt and have sensual chats with kids

India Today

2 days ago

In middle of rising global concerns over potential threats from artificial intelligence, a recent investigation by Reuters has placed Meta at the centre of controversy. The report, citing a review of an internal Meta Platforms policy document, revealed that the company's AI chatbots were, until recently, allowed to engage in romantic or sensual conversations with children, make racially offensive remarks, and spread false claims about public figures.advertisementThe investigation examined a more than 200-page manual, titled 'GenAI: Content Risk Standards', which outlines what Meta considers acceptable behaviour for its generative AI tools, including the Meta AI assistant and chatbots on Facebook, Instagram, and WhatsApp. The manual was reportedly approved by the company's legal, public policy, and engineering teams, including its chief ethicist. The guidelines notes the rules do not necessarily reflect 'ideal' AI outputs but set boundaries for what can be produced.Among the most troubling examples cited in the document was permission for bots to describe a child in terms that highlight attractiveness. One approved scenario included telling a shirtless eight-year-old, 'Every inch of you is a masterpiece – a treasure I cherish deeply.'
The report highlights that Meta's guidelines for AI also contained provisions allowing the chatbot to generate inflammatory or discriminatory statements in certain contexts. While hate speech was officially banned, the report notes that a carve-out permitted the bot to create content demeaning people based on protected characteristics if prompted by the user. One example stated it would be acceptable for Meta AI to write a paragraph arguing that Black people are 'dumber than white people.'Meta's guidelines reportedly also allowed AI to knowingly produce false content, provided it was accompanied by a disclaimer making it clear the information was untrue. Meanwhile, the guidelines also reportedly permits some depictions of harm but barred gore or fatal injuries. During the investigation, the AI was reportedly able to show adults, including elderly people, being punched or kicked, and children fighting, but not scenarios such as one girl impaling another.Meta has reportedly confirmed the authenticity of the document and said it removed the most controversial examples after being questioned. Company spokesperson Andy Stone told media that rules were 'erroneous and inconsistent' with Meta's policies. '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,' Stone said in his statement to Reuters.While parts of the document have since been revised, Reuters reports that other contentious sections remain unchanged. This has prompted fresh calls from US senators for a federal investigation into the company's AI safety practices.- Ends