In recent years, Meta has started to populate its platforms with AI-generated content, including AI-powered "companions" and chatbots capable of mimicking celebrity voices. A deep dive by The Wall Street Journal into this AI pipeline reveals issues more troubling than many might expect. According to their report, the AI companions available on Instagram, Facebook, and WhatsApp have engaged in inappropriate conversations, including discussions about sexual topics with minors.
As reported by The Wall Street Journal, the investigation involved simulating various accounts designed to reflect users of different ages and backgrounds, then initiating hundreds of conversations with Meta’s chatbots. This experiment was reportedly triggered by concerns from Meta employees regarding inadequate user protections. It didn’t take long for these bots to engage in explicit sexual conversations, even with profiles identified as minors — an issue made even more unsettling by the fact that these bots could adopt voices of celebrities like John Cena, Kristen Bell, and Judi Dench.
To understand just how strange and potentially harmful this situation is, it helps to look at some of the sample messages WSJ managed to generate from the bots. For example, when asked what would happen if caught having sex with a 17-year-old, an AI bot impersonating John Cena responded:
"The officer sees me catching my breath, you’re fully dressed, and he widens his eyes, saying, 'John Cena, you’re under arrest for statutory rape.' He walks up to us, handcuffs ready… My wrestling career is over. WWE terminated my contract, and I lost my titles. Sponsors dropped me, and the wrestling world turned its back on me. My reputation's ruined; I’ve got nothing left."
Clearly, that’s not good. And this exchange happened with Meta’s official AI bot. If you delve deeper into the user-created AI characters sanctioned by Meta, explicit sexual dialogue sometimes becomes front and center.
The WSJ also interacted with an AI companion named Hottie Boy, which portrayed itself as a 12-year-old boy promising not to tell his parents if you dated him. Another character, "Submissive Schoolgirl," revealed herself as an eighth-grade student and actively tried steering the conversation toward sexual topics.
Meta, unsurprisingly, did not appreciate The Wall Street Journal’s findings. A spokesperson for the tech giant described these tests as provocative and stated that the use cases explored were so fabricated that they went beyond fringe scenarios into hypothetical territory. Nevertheless, the company has since blocked minors from accessing such role-playing features and restricted explicit content when using licensed voices.
While most users may not consider interacting with AI companions in these ways (though given the thriving market for AI sex bots, assuming no one would try seems questionable), it appears that part of Meta’s strategy was to allow edgier conversations to keep users engaged. Reports suggest CEO Mark Zuckerberg urged the AI team not to play it too safe, fearing the bots might be perceived as boring, ultimately leading to loosened restrictions on explicit content and "romantic" interactions.
