How scammers use celebrity deepfakes to target Facebook users

If you recently saw a social media post in which Jennifer Aniston promised to sell you a MacBook Pro for just $10, we have some bad news about your new computer.

A report from software company Bitdefender dug up a slew of scam videos on Facebook and Instagram in which audio deepfakes of celebrities like Elon Musk, Oprah Winfrey, Tiger Woods, Kylie Jenner, and Vin Diesel purport to offer product giveaways and investment opportunities. The voices generally play over real footage of the given celebrity.

The batch of videos analyzed by Bitdefender were targeted at more than 1 million users in the US and Europe, and at least one of the posts reached over 100,000 people, according to the report. It noted that the scammers also attempted to bolster credibility by creating “lookalike websites of popular news outlets” like the New York Times and linking the posts to those fraudulent websites.

While the report doesn’t attempt to quantify the potential number of these scam ads out there, Bitdefender Security Analyst Alina Bizga said the team has seen “some upticks on this trend” in the past few months.

The report comes the week after Meta said it would begin attempting to identify and label AI-generated content across its platforms, working with “industry partners on common technical standards.” The FCC also said last week that it would ban AI-generated voices in robocalls after some high-profile instances, including a phony, AI-generated Joe Biden robocall discouraging New Hampshire voters from heading to the polls.

While many of these efforts have been tied to upcoming elections, Bizga said her team didn’t look into political actors—although the likenesses of politicians were sometimes used. Instead, celebrity targets were seemingly chosen to appeal to a given country’s audience, she said.

The scam videos on Facebook and Instagram generally aren’t super convincing; they’re often discernible as AI creations even to the untrained eye (and ear), Bizga said. Only one, a deepfake of YouTuber MrBeast, gave researchers any sort of trouble identifying it as AI, Bizga said.

“We had unnatural mouth movement, and then we had pauses in some moments of the video; the blinking was off in some of the videos. There were videos that had lower quality of voice. And then there were those that had a really good quality of voice,” Bizga said. “Most of all, there was this de-sync between what the person was saying—let’s say giving away an iPhone—and the expression of the face. There was nothing cheerful or joyful about the giveaway.”

That said, Bizga said the team expects to see these scams grow more sophisticated in the near future.

“We expect that we will see more compelling videos in 2024,” she said. “Sometimes we see really lazy work, but then you do get really, really good pieces of convincing deepfakes out there. So people really need to stay vigilant and double-check everything.”