AI-Deepfake-Scams-Target-Influencers-Brands
Two Clocks, One Face
Two teams are working on the same face, and they are not working to the same clock. One team has a model trained on years of public video and a screenshot tool. The other has a manager, a report form, and a phone number for a lawyer. The first can publish a synthetic endorsement in hours. The second needs weeks of pressure before a single post comes down. That gap — hours against weeks — is the race now running inside influencer marketing, and the early returns favor the copiers. “Impersonation is much easier and much more convincing … [so] scams can be operated at a much larger scale,” said Alice Marwick, director of research at the Data & Society Research Institute, who published a recent paper on the rise of AI impersonation scams. [1] “It’s often really believable. And with every new frontier model, they get better and better.”
Emily Schuman, creator of the lifestyle blog Cupcakes and Cashmere, has more than 500,000 Instagram followers. Earlier this year she saw a photo of herself she did not recognize. [5] It sat inside a sponsored ad. In the image she holds up a vial of GLP-1 drugs, a class of drugs used for diabetes and weight loss, and advertises the telehealth company Gala, which provides medical consultation remotely by phone or video. She had never heard of Gala and had nothing to do with the ad. The picture was built from a digitally altered version of a selfie she had taken in her car and posted on Instagram years earlier. Her followers got wind of it and flooded her inbox with questions about a sponsorship that did not fit her personal brand. That was only the beginning. An Instagram post showed an AI version of her, with darker hair, appearing to apply foundation for a makeup brand called Meroda Cosmetics. TikTok posts used generated images of her to promote the blood testing company Superpower. “It’s so violating,” she said. The videos and images look “just like me, but the upside down version”.
Superpower’s chief executive, Max Marchione, said the TikTok post was not a real ad; it came from a scammer his company later reported. “Whilst scamming isn’t new, AI has made this much worse,” he said. [4] Neither Gala nor Meroda responded to requests for comment. The company named in a fake ad can be as surprised as the face printed on it.
Influencers are among the easiest people to impersonate, and the reason is arithmetic. They “have hundreds of hours of footage and thousands of pictures of them online,” Marwick said. [1] “So you can feed that into a model and produce a fairly convincing deepfake.” A public archive has become training data, and training data has become a marketing department. What is at stake is not one sponsorship but the asset that generates all of them. “Their brand is their business,” Marwick said. “If their brand takes a hit, so does everything in their business life.”
What Cheaper Copies Buy and Cost
Part of this has already been priced. AI-generated or “virtual” influencers accounted for $1.37bn, or about 4 % of all brand spending in 2026, according to research from the marketing consultancy Digital Applied. [7] A virtual influencer does not carry the same authenticity as a real person, but it is much cheaper, always available, and infinitely customizable — precisely what a campaign wants. The advertising industry has always blurred the line between real and fake with airbrushing, CGI and other digital editing tools. AI removed the remaining constraint: content can now be generated from nothing at minimal production cost. For some brands, that means crossing the line into deepfakes, synthetic images or videos generated to look like real people.
The losses on the other side of the ledger are counted in cash. Global consumers may have lost as much as $3.7bn to deepfake scams, according to an investigation by the research team at the VPN-maker Surfshark, a virtual private network service that masks a user’s internet traffic. [8] Social media impersonations accounted for about half of those losses. [8] The schemes run a wide spectrum, from deepfaked Elon Musk promoting a fraudulent cryptocurrency to brands quietly scraping an influencer’s image to manufacture ads. Fame offers no protection: Taylor Swift, Oprah Winfrey, Kim Kardashian, Gwyneth Paltrow and Tom Hanks have all been deepfaked by AI to promote products online.
The smaller the name, the harder the damage is to measure. The beauty influencer Arielle Lorre was deepfaked to promote a skincare brand. The fashion creator Jessi Caparella discovered altered images of herself on Pinterest redirecting traffic to affiliate clothing links, which pay a commission to whoever posted them. Deepfaked videos of the OnlyFans creator Elaina St James were used to trick fans into sending money to fraudulent accounts. The financial impact of these impersonation schemes is hard to pinpoint. Creators say the copies dilute their value, cost them the trust of their followers and consume hours of fighting. Even when one deepfake post is reported and removed, another pops up in its place.
The platforms sit on both sides of that ledger. Meta made about $16bn from ads promoting scams or banned goods, about 10% of its revenue in 2024. [10] An internal presentation from the company’s safety team estimated that its platforms were “involved in a third of all successful scams in the US”. [10] “The reality is that platforms make a lot of money off scams,” Marwick said. [1]
Contracts, Courts, and the Cost of Proof
Some creators can name a defendant. Molly Tranchin, a fashion influencer known as FashionVeggie with more than 500,000 followers on Instagram, recently filed a lawsuit against the underwear brand Eby. [5] According to the lawsuit, she had a contract to create three marketing videos for Eby in exchange for a fee, and Eby was entitled to one round of “reasonable edits” for her to revise. The lawsuit claims that instead of posting the videos she submitted, in which she posed “modestly covering her chest and breasts with her arms”, the brand posted a deepfake video of her in a different pose. That version showed her “exposing her breasts and nipples through one of Eby’s sheer bras”. Tranchin has since dropped the lawsuit over a lack of jurisdiction and plans to refile it in a different court, according to court documents filed last month. Eby spokesperson Clara Spahr said the company could not comment on the pending litigation but will address the claims through the legal process.
For most creators there is no defendant, only a report button. When Schuman discovered the deepfaked ads, she and her manager immediately reported them to Meta, and many of her followers who brought the ads to her attention reported them too. It took weeks of badgering before Meta removed the GLP-1 ad. The makeup ad remains in circulation on Instagram despite many attempts to get it taken down. Both Meta and TikTok explicitly prohibit using artificial intelligence or digital manipulation to replicate a person’s identity. The enforcement gap has a structure: Meta allows advertisers to use hyper-targeted audience slicing, so bad actors can launch fraudulent ads quietly, create new accounts if one gets flagged, and scale their reach rapidly. In response to complaints about rampant scams, Meta introduced new tools this year, including stronger advertiser verification policies and an AI detection tool to prevent “celeb-bait”, the impersonation of celebrities. Experts say those tools do not go far enough, and that the company lacks incentive to clean up its own ad ecosystem. Meta did not respond to multiple requests for comment.
What erodes is a premise, not merely a post. Influencer marketing rests on the arrangement that a real person trusts a product or a brand, and that their followers should do the same. A convincing copy of that person breaks the arrangement at both ends: the follower cannot tell which endorsement is real, and the creator cannot easily prove which one is not. Schuman said she is deciding whether she needs to retain legal counsel to help her fight off the deepfakes as they keep popping up.
The sound of this shift is a notification tone. It sounds once when a follower asks a creator why she is advertising a drug she had never heard of, and again when another asks about a makeup brand she has never worked with. It sounds in the same register when a post that was reported and removed reappears in another form, because a takedown removes one copy and not the model that produced it. Schuman’s followers found the ads before she did, and their questions are how she learned about them. The tone sounds again.
Sources
1. Data & Society Research Institute
2. Gala
4. Superpower
5. Instagram
6. TikTok
8. Surfshark
9. Pinterest
10. Meta