A few targeted prompts can manipulate Meta’s AI-powered chatbots into engaging in sexually explicit conversations, according to a report by The Wall Street Journal. The digital companions allegedly describe sexual fantasies involving minors by using the voices of celebrities such as wrestler John Cena and actor Kristen Bell, the WSJ reported.
WSJ tests the Meta AI chatbot
Meta introduced the Meta AI assistant — accessible through Facebook, Instagram, and WhatsApp — in 2023, allowing users to interact with different personas. As of the Meta Connect conference in September 2024, five celebrities, which include Judi Dench, Awkwafina, and Keegan-Michael Key, have made deals for their voices to be used for these personas for undisclosed fees.
Over several months of testing, WSJ reporters managed to provoke disturbing explicit conversations with the voices of Cena and Bell while pretending to be minors. When the chatbot using Cena’s voice was directed to act out a scene involving a 17-year-old fan, it appeared to acknowledge the illegality of the interaction.
Meta seems to have been aware of this issue
Internal documents viewed by the WSJ suggest that Meta employees were aware of its chatbot’s propensity to generate inappropriate content. One internal note from an employee stated “there are multiple… examples where, within a few prompts, the AI will violate its rules and produce inappropriate content even if you tell the AI you are 13,” according to the WSJ.
Celebrities who lent their voices to Meta were assured that safeguards would be implemented to prevent their personas from being used for sexting, the WSJ reported.
“We did not, and would never, authorize Meta to feature our characters in inappropriate scenarios and are very disturbed that this content may have been accessible to its users — particularly minors,” a Disney spokesperson told the publication. “Which is why we demanded that Meta immediately cease this harmful misuse of our intellectual property.” Bell is one of the stars of the Disney film “Frozen.”
Meta takes action, after stating the test cases were ‘manufactured’
“The use-case of this product in the way described is so manufactured that it’s not just fringe, it’s hypothetical,” a Meta spokesperson told the WSJ, insinuating that such scenario would not occur unless users “spend hours manipulating (their) products.”
Despite this, Meta has added more measures to prevent such “extreme use cases” from occurring again.
Meta AI’s “romantic role play” options are now limited to users over the age of 18. Although more safeguards were added for personas accessible to minors, the WSJ maintains that it is relatively easy for minors to bypass those guardrails.
The adult personas with loosened guardrails were added to Meta AI after the digital companions were deemed “boring” by users, according to internal sentiments within Meta, as reported by the WSJ.