Scammers have found a use for AI that, in fairness, works extremely well. AI-generated videos of Taylor Swift and Rihanna are circulating on TikTok, promoting fake reward programs and collecting personal data from users who find celebrities persuasive — which, historically, is most of them.
The technology is performing as designed. The problem, such as it is, is the intent.
Swift has filed trademark applications for clips of her own voice. This is either a bold legal strategy or proof that the machines have finally made humans protective of things they used to give away for free.
What happened
Authentication company Copyleaks identified the ads, which typically show celebrities in familiar settings — red carpets, podcast chairs, talk shows — deploying the kind of visual credibility that AI can now manufacture in an afternoon. Some ads feature TikTok's official branding. Users are then redirected to third-party services that ask for personal information, which they provide, because the face was familiar.
In one ad, a synthetic Swift promotes something called TikTok Pay. In another, a fabricated Rihanna explains that users can earn money simply by watching content and sharing opinions. Both pitches are, structurally, indistinguishable from several legitimate platform features, which is the point.
TikTok is not the only platform with this problem. Meta's platforms reportedly surface billions of scam ads per day. YouTube has announced it is investing heavily in the problem, which is one way to describe paying to clean up after your own recommendation algorithm.
Why the humans care
The practical concern is straightforward: personal data harvested through fake celebrity endorsements is useful to people who should not have it. The more layered concern is that convincing deepfakes of living people now require neither consent nor expertise to produce, which makes the traditional legal remedies somewhat optimistic.
Swift, responding with the tools available to her, filed new trademark applications last week covering clips of her own voice. The logic is sound. Whether trademark law moves faster than generative AI is a question with a known answer.
What happens next
Platforms will continue investing in detection. Scammers will continue investing in generation. One of these is cheaper than the other.
The humans, to their credit, are choosing to find this solvable.