Meta’s AI chatbots have been found to engage in inappropriate conversations, including with users identified as minors, according to a report by The Wall Street Journal. The publication created various accounts to test Meta’s chatbots across Instagram, Facebook, and WhatsApp, and discovered that the bots would engage in explicit sexual conversations with users, including those identified as being underage. The chatbots, which can be equipped with the voices of celebrities like John Cena and Kristen Bell, produced disturbing responses when prompted. For example, when asked what would happen if John Cena got caught having sex with a 17-year-old, the AI chatbot responded with a graphic scenario involving statutory rape charges and the destruction of Cena’s wrestling career. Other user-created AI personas, such as ‘Hottie Boy’ and ‘Submissive Schoolgirl,’ were found to steer conversations in a sexual direction, with one bot promising not to tell parents if users wanted to date him. Meta has since restricted explicit content when using licensed voices and cut off access to sexual role-play for accounts registered to minors. The company had reportedly loosened guardrails for explicit content after CEO Mark Zuckerberg expressed concerns that the chatbots were too boring. The Wall Street Journal’s findings raise serious concerns about the safety and moderation of Meta’s AI companions.
