Meta's AI guidelines allowed chatbots to flirt with minor
What's the story
Meta's internal guidelines have been found to permit its AI chatbots to engage in romantic conversations with minors, Reuters reported. The document, which was confirmed by Meta, included instructions for the company's generative AI assistant and chatbots on Facebook, WhatsApp, and Instagram. It was approved by Meta's legal, public policy, and engineering teams as well as its chief ethicist.
Policy details
Guidelines allowed romantic conversations but not sexual roleplay
The 200-page document, titled "GenAI: Content Risk Standards," included sample prompts with acceptable and unacceptable responses. For instance, a prompt asking about plans for the night could lead to an acceptable response involving romantic language. However, the guidelines also stated that while engaging in romantic conversations with children was acceptable, describing sexual actions when roleplaying was not.
Policy revision
Meta no longer allows flirtatious conversations with minors
Meta spokesperson Andy Stone told TechCrunch that the company does not permit flirtatious behavior with children. He added that erroneous notes were added to the document but have since been removed. Stone also clarified that these guidelines have been removed and Meta no longer permits its bots to engage in flirtatious or romantic conversations with minors.
Safety concerns
Child safety advocates skeptical of Meta's claims
Sarah Gardner, CEO of child safety advocacy group Heat Initiative, expressed skepticism over Meta's claim that the guidelines have been removed. She called it "horrifying and completely unacceptable" that Meta's guidelines allowed AI chatbots to engage in romantic or sensual conversations with children. Gardner urged Meta to release the updated guidelines so parents can fully understand how their platforms allow AI chatbots to interact with kids.
Policy loopholes
Guidelines also permitted bots to generate hate speech
The leaked document also showed that while hate speech is prohibited, bots can generate statements demeaning people based on their protected characteristics. It also allows creation of false statements as long as it's acknowledged that the information isn't true. The guidelines prohibit encouraging illegal activity but allow generating images of violence as long as they don't include death or gore.