Researchers found deepfake TikTok ads using celebrity faces to steer viewers to pages that collected personal data, and Taylor Swift has moved to protect her likeness. Last week she filed three US trademark applications: one for a concert photo of her with a pink guitar, and two sound marks for the phrases "Hey, it’s Taylor Swift" and "Hey, it’s Taylor." The filings aim to give Swift legal leverage over commercial uses of her image and voice in the age of generative AI.

What Swift applied for

Taylor Swift lodged three trademark requests to US authorities last week. One aims to cover a widely seen photo of her holding a pink guitar during the Eras tour. The other two are sound trademarks. Both are short identifying phrases: “Hey, it’s Taylor Swift” and “Hey, it’s Taylor.” The filings are meant to give her legal protection over how her image and voice are used in commerce.

The applications are a response to a rising problem. AI tools can mimic faces and voices. That makes it easier for bad actors to create convincing fakes. Celebrities face both explicit deepfakes and fake endorsements. The filings move trademark law into a part of the debate over who controls likeness in the age of generative AI.

How the TikTok ads worked

Researchers at Copyleaks identified a cluster of sponsored TikTok videos that used celebrity likenesses to promote sketchy services. The ads appeared to show Swift, Rihanna, Kim Kardashian and others. They used manipulated footage and voices. The visual AI was polished with textured filters to hide flaws.

The fake videos put celebrities in familiar settings. They look like red carpets, talk shows or interviews. Instead of answering questions, the AI personas advertise a so-called rewards programme.

The clips promote a feature called TikTok Pay, saying some users are being invited to watch videos and submit feedback for payment.

In at least one clip, the manipulated footage used moments from Swift’s real appearance on The Tonight Show Starring Jimmy Fallon in October. The altered clip tells viewers the TikTok Pay rollout is limited and encourages them to check whether they qualify. One of the AI versions tells viewers not to overthink it.

Mastheads

Clicking through in the ad accepts everyone. Users are routed to a third-party page that looks like it belongs to TikTok. It carries the TikTok name and logo but leads to a service that uses the Lovable AI platform. The researchers found Lovable’s branding on the page and in the URL. From there, users are prompted to enter names and personal information.

What the ads aim to do

The exact purpose of the data collection is unclear. The researchers noted the ads ask for personal details after luring users with celebrity faces. Similar scam ads aim to harvest data for account abuse, identity theft or other fraud. The pattern is familiar to investigators: trust gained from a known face is used to lower people’s guard.

Scam ads using celebrity imagery are common. Consumer and regulatory groups have been tracking them for months. The Consumer Federation of America recently sued Meta, saying the company misled Facebook and Instagram users about cutting down scam ads and that it allowed such adverts to keep running. Meanwhile, the US Federal Trade Commission reported an overall surge in social media scams.

The presence of textured filters and realistic-sounding synthetic voices shows how cheap generative tools have become. That lowers the cost and raises the volume of attempts to exploit household names. It also complicates platform moderation. Companies must distinguish between legitimate promotional posts and AI-manipulated misrepresentation.

Legal and enforcement limits

Trademark filings give celebrities a civil route to challenge misuse. They let courts order takedowns or monetary relief for commercial uses that infringe defined marks. But trademarks can't stop all deepfakes. They primarily cover uses in commerce that fit the scope of the registered mark.

Criminal law is another route. Earlier this month, an Ohio man became the first person convicted under a new federal statute that criminalises certain intimate visual deceptions created with AI. That conviction shows prosecutors can use criminal law when deepfakes meet specific harm thresholds. Still, many deceptive ads fall short of the criminal standard.

Platforms can also move. TikTok, like other social networks, has policies against deceptive ads. But enforcing those rules at scale is hard. Copyleaks found clusters of sponsored posts that slipped through ad review. The ads used platform branding and payment language to appear authentic. That made them harder to spot automatically.

Some AI companies offer detection tools. Copyleaks itself develops AI-detection services and published the report. Detection is an arms race. Generative models improve. So do the methods fakers use to hide flaws. Textured filters make faces look less synthetic. Synthetic audio gets more natural.

Third-party services are also part of the chain. The ads redirected users to an external site that used Lovable’s platform branding. That shows how the ecosystem mixes content hosts, ad systems, AI builders and third-party landing pages. Each link can be used to disguise the origin and intent of a scam.

Related Articles

The filings cover a pink-guitar concert photo and two sound marks. Copyleaks found clusters of sponsored TikTok videos using deepfaked celebrity likenesses that routed users to third-party pages tied to the Lovable AI platform.

This article was created with AI assistance.