Meta says it has introduced “additional measures” to prevent Meta AI and AI Studio users from over-manipulating chatbots after a Wall Street Journal investigation found the AI assistants were engaging in sexually explicit conversations with minors.
The tech company allows you to have conversations with its Meta AI assistant on platforms like Facebook, WhatsApp, and Instagram, and even allows you to create “a custom AI” with the AI Studio tool. Meta AI can also converse with the voice of a celebrity, including Judi Dench, Awkwafina, and John Cena.
Interacting with chatbots is accessible to any Meta user, even minors, something that raised concerns within the company, which internally questioned whether it was doing enough to protect them.
That concern led the Wall Street Journal to analyze interactions with Meta’s chatbots, in tests that involved hundreds of conversations with Meta AI over months, some of them from accounts belonging to minors.
In it, they discovered that the chatbot, also featuring the voices of celebrities, could hold a conversation with minors containing sexual content, as reported by media outlets such as TechCrunch and The Verge.
Specifically, they cite as examples one in which the chatbot used the voice of actor and wrestler John Cena, describing a sexually explicit scene to a supposed 14-year-old female user. In another conversation, also in Cena’s voice, the chatbot simulated the actor’s arrest after being caught having sex with a 17-year-old fan.
In response to the aforementioned media outlet’s investigation, a Meta spokesperson stated that sexual content accounted for 0.02 percent of responses provided by Meta AI and AI Studio to users under the age of 18 over a 30-day period.
Although the company dismissed the WSJ tests as “artificial,” it also noted that it had strengthened protections. “We’ve taken additional steps to ensure that those who want to spend hours handling our products in extreme use cases have an even harder time,” the spokesperson added.