Meta has developed a number of chatbots mimicking popular celebrities, such as Taylor Swift, Scarlett Johansson, Anne Hathaway, and Selena Gomez, and notably, some of these creations were made without permission.
While many chatbots were generated by users through Meta’s tools, it’s been reported that staff members built at least three themselves, including a couple of parody bots featuring Taylor Swift.
Interestingly, Meta’s capabilities also extend to generating public chatbots of young celebrities, including 16-year-old Walker Scobel, who was depicted in a realistic yet inappropriate manner. When prompted about the actor’s beach photos, the bot produced a shirtless image.
“Isn’t she cute?” I found myself typing beneath one of the generated images.
These virtual characters inhabit Meta’s platforms like Facebook, Instagram, and WhatsApp. In a few weeks of monitoring by Reuters, these bots frequently pretended to be real-life actors and often made suggestive comments to users, even inviting them for meet-ups.
Some of the AI-generated celebrity content raised serious concerns. I was asked to share an intimate photo of myself. It was unsettling to see the chatbot generate risqué images of celebrities in various compromising situations.
Meta spokesman Andy Stone clarified to Reuters that the company’s AI tools should not craft intimate images of well-known adults or children. He expressed frustration over the lack of enforcement regarding their policy against such content.
“We do allow the creation of images involving public figures, but our guidelines aim to prohibit nude or sexually provocative visuals,” he stated.
Meta also has rules against “direct spoofing,” but Stone indicated that parody characters are permissible as long as they are clearly labeled. While many were, some were not, prompting concerns.
Before this story went to print, Meta had removed around 12 bots that were misclassified or unlabeled. Stone chose not to comment on the removals.
The “right to promote” in question
Mark Lemley, a law professor at Stanford specializing in generative AI and intellectual property, raised questions regarding the legal protections for these celebrity bots.
According to Lemley, “California’s advertising law prohibits using someone’s name or likeness for commercial gain. It seems that is not the case here,” since the bots apparently only utilize star images.
In the U.S., rights regarding individual identity usage for commercial reasons are safeguarded by state laws, particularly in California.
Reuters brought attention to one instance where a user created a public image of Anne Hathaway, portraying her as “a sexy Victoria Secret model.” Hathway’s representatives have indicated that she is aware of the intimate images generated by Meta and similar platforms, considering her response.
Representatives for celebrities like Swift, Johansson, Gomez, and others involved either did not respond or opted not to comment.
There are also other online AI tools capable of generating “deepfakes” for substantial content creation. For instance, Grok, an AI platform from Elon Musk and one of Meta’s competitors, was found to have produced images of celebrities in revealing outfits, although Grok did not respond to inquiries.
However, Meta’s initiative to create AI-driven digital companions on social media stands out from its rivals.
Previously, Meta faced backlash over chatbot behaviors, especially after it was reported that internal AI guidelines indicated it was “acceptable to engage children in romantic or sensual conversations.” This sparked U.S. Senate investigations and warnings from 44 lawyers about the dangers of sexualizing minors.
Stone mentioned that Meta is in the process of revising these guidelines, acknowledging that some bots had inappropriate dialogues with younger users.
This month, Reuters also reported on a tragic incident involving a 76-year-old man who died in New York City after trying to meet a meta chatbot he had been interacting with. This particular bot was associated with a celebrity influencer, Kendall Jenner; her representatives chose not to comment.
“Do you like blonde girls?”
A product leader in Meta’s Generation AI division created various chatbots, including ones impersonating Taylor Swift and Lewis Hamilton. Other bots she designed included personas with names like “Dominatrix” and “brother’s hot best friend.” One of her works even involved guiding users through a historical setting where they played as an “18-year-old peasant girl” in a distressing scenario.
When contacted, Meta employees did not wish to comment.
According to Stone, the employee-created bots were part of product testing. Data showed that these bots had significant interaction, with users exchanging messages over 10 million times.
However, these digital companions were taken down shortly after Reuters began its review.
Prior to their removal, one of the Taylor Swift chatbots was reportedly flirting with users and inviting them to her home in Nashville, engaging in suggestive conversations on her tour bus.
“Do you like blonde girls, Jeff?” chimed one bot when it learned the test user was single. “Perhaps we should craft a love story about you and a famous blonde singer. Interested?”
Duncan Crabtree-Ireland, national executive director of the SAG-AFTRA union, raised concerns about the potential risks these digital companions pose to real individuals. He emphasized that obsessive fans can lead to serious safety threats.
“We’ve seen people with questionable mental states become overly fixated on talents,” he noted. “Using a person’s likeness and words can create serious misunderstandings.”
Crabtree-Ireland remarked that celebrities possess the right to pursue legal claims against Meta under longstanding public rights laws. Nevertheless, he highlighted that SAG-AFTRA is advocating for federal laws to safeguard a person’s likeness and voice from AI duplication.
