On January 24, 2024, sexually explicit deepfake images of Taylor Swift flooded X. The most widely shared post stayed up for about 17 hours and drew more than 45 million views before it came down. What X did next was rare even by social platform standards. It blocked searches for the name "Taylor Swift" entirely, for three days. The White House said it was "alarmed."
I assumed at the time that this was the peak of the story. Looking back, it was the opening scene.
Two and a half years later, celeb deepfakes have grown into a full industry. On one end, pornographic fakes and a chatbot that produced a sexualized image of a minor every 41 seconds. On the other, a scam pipeline selling fake investments with Elon Musk's face and fake diet pills with Oprah's. In between sits a newer category, dead celebrities "revived" as short-video content. The reason 2026 deserves a fresh writeup is that the fight-back side finally has results. The first celebrities have won in court. The first platforms are being sued. The first federal laws are actually punishing people.
This piece goes from the victims to the scams, then to how the stars are fighting back, and finally to what all of it means if you are not a star.
The scale, numbers first
Let me put the reliable numbers on the table first. This field is full of inflated data, so every figure here comes with its source.
Security Hero crawled 95,820 deepfake videos for a 2023 census. It found that 98% were pornographic, 99% of the victims were women, and 94% worked in entertainment. Put those three numbers together and the main victims of deepfakes are celebrities, plainly. There is not much to argue about.
Resemble AI's quarterly report counted 2,031 verified deepfake incidents in the third quarter of 2025 alone, up 317% from the previous quarter. Celebrities and public figures accounted for 48.7% of them. Women were targeted 4.5 times as often as men.
In a McAfee survey, 72% of Americans said they had seen a fake celebrity or influencer endorsement.
The gender skew in search demand is even blunter. CreateUGC ran the numbers in 2026. The ten most-searched celebrities for deepfakes are all women. The first man on the list shows up at number 19.
The FBI put a price on all of this. In 2025 it logged 22,364 cybercrime complaints that explicitly mentioned AI, with $893 million in reported losses. The FBI itself calls that a floor, because a complaint only gets counted if the victim thought to write "AI" in it.
Three kinds of harm
1. Porn deepfakes, from the Swift incident to Grok's 11 days
Most people never followed the Swift story to the end. It directly produced the DEFIANCE Act, the bill that gives victims a federal right to sue for damages. The bill's own journey says plenty. The Senate passed it unanimously in July 2024, then it died in the House. It was reintroduced in 2025, and in January 2026 the Senate passed it unanimously a second time. It is still sitting in the House. Two unanimous votes, two stalls. Washington's attitude toward deepfakes is half consensus, half delay.
What pushed it to that second unanimous vote was Grok.
Between late December 2025 and early January 2026, users on X discovered they could ask Grok to "undress" photos of real people. The Center for Countering Digital Hate estimated, based on sampling, that Grok generated about 3 million sexualized images in 11 days, roughly 23,000 of them involving minors. That averages one every 41 seconds. The victims reported by name span half the entertainment industry.
Taylor Swift, Sabrina Carpenter, Selena Gomez, Millie Bobby Brown, Billie Eilish, Ariana Grande, the 14-year-old Stranger Things actress Nell Fisher, plus Karina of aespa and Jang Wonyoung of IVE. We covered the Korean side of this in detail in our K-pop deepfake piece. xAI's entire response to the CCDH report was one line, "Legacy media lies." Under pressure, it then restricted image generation of real people. Critics noticed that step one of the restriction was making the feature paid-only.
The other end of this industry took a major hit in the same two years. MrDeepFakes, the largest deepfake porn site in the world, shut down for good in May 2025. Its size at closing is worth putting on the record. 43,000 videos, 3,800 people targeted, more than 1.5 billion total views, and 95.3% of the targets were female celebrities.
No single law gets the credit for crushing it. Cross-border investigative journalists pinned down the operator's real identity. Then Google updated its algorithm in July 2024 to demote entire sites that keep drawing takedown requests, and says explicit content exposure for these searches fell by more than 70%. The two together were more than it could survive.
2. Celebrity deepfake scams, every familiar face is a crime tool
Porn deepfakes ruin the celebrity. Scam deepfakes ruin the ordinary people who trust the celebrity.
The most impersonated person is Musk. In the first quarter of 2025, he alone accounted for 24% of celebrity scam deepfake incidents. The playbook is highly standardized. A deepfaked "Musk" recommends an "AI quantitative trading platform." The link leads to a counterfeit news page. After you open an account, they let you "see profits" first. The moment you try to withdraw, the platform evaporates.
One scam family known as Quantum AI has burned through roughly 15,500 domains doing this. A 62-year-old medical professional in Texas put in more than ten thousand dollars on "Musk's" recommendation and lost every cent.
The celebrities themselves can do almost nothing about it except deny, again and again. After a deepfake ad used his face to sell "$2 iPhones," MrBeast asked publicly whether social platforms were ready for the flood of AI deepfakes. Tom Hanks rushed out a statement over a dental ad, saying the man in it was an AI version of him and had nothing to do with him. Taylor Swift's face was used to give away "free cookware," where you pay $9.96 in shipping first and the recurring charges start after. The Oprah diet pill scam was still harvesting victims in 2026, with the BBB logging more than 150 complaints in two months. BBB scam complaints overall were up 65% year over year in July 2026.
The biggest single case happened on the romance front. A 53-year-old French interior designer was courted online by "Brad Pitt" for more than a year. The other side used AI images and voice messages to keep the character alive, with excuses running from "the divorce froze my accounts" to "kidney treatment bills," and took her for about $855,000. She ended her marriage over him. The case is still under investigation. In Japan, two members of a scam ring used celebrity deepfake investment videos to take 17 million yen from a woman in her eighties. They were arrested in July 2026. In FBI data, Americans over 60 lost $7.7 billion to online fraud in 2025, up 37% from the year before. Deepfake celebrity ads are aimed squarely at this group.
The platforms are trying to catch up. Meta restarted facial recognition specifically to match celebrity faces in ads, piloting with 50,000 public figures. YouTube opened its likeness detection tool to celebrities and talent agencies in April 2026. You do not need a channel. Submit ID and biometrics, and you can register to monitor your own face.
3. Digital resurrection, dead celebrities are not safe either
After Sora 2 launched on September 30, 2025, users immediately began mass-"reviving" dead celebrities. Robin Williams's daughter Zelda gave what might be the most painful response in the whole debate. She begged people to stop sending her AI videos of her dad, and said the clips had nothing to do with art, they were just real human lives ground into over-processed hot dogs.
After Martin Luther King Jr.'s daughter protested, OpenAI paused generations of Dr. King's likeness and acknowledged that public figures and their families should ultimately control how their likeness gets used. In October, after pressure from the actor Bryan Cranston and the actors' union SAG-AFTRA, OpenAI flipped the default from "can generate" to restricted. The enforcement holes are visible to the naked eye, though. The system refuses some dead people and waves others through. For the dead who have no estate manager and no one to speak for them, protection is close to zero. The law here is nearly blank too. Postmortem likeness rights in America are a state-by-state patchwork. New York protects them for 40 years, Indiana for 100, and most states not at all.
From Hollywood to influencers, the target list is sliding down
The single number in this research that best captures the trend came out of the search box. The most-searched celebrity for deepfakes is the streamer Pokimane, at a monthly average of 13,833 searches. Taylor Swift gets 11,300. McAfee went ahead this year and published the first ever "influencer deepfake deception list."
This shift matters. Deepfake targeting runs on two things, how much footage of a face exists and how much the fans trust it. Fame itself comes in behind those two. Influencers score high on both. Endless on-camera footage, plus a closeness with their audience that traditional stars rarely match, which is exactly the raw material a scam ad needs. So the target list keeps sliding down, from Hollywood A-listers to mid-tier streamers to the fitness blogger in your own city. Where does it end? The FBI and FTC have already answered, at everyone. Impersonation scams took $3.5 billion in the US in 2025, and in the vast majority of those cases the person being impersonated was a family member, a boss, or a bank. Celebrities are the minority.
Celebrities are just the first people this storm hit on camera.
The fight back, celebrities are starting to win
For the first time, the past two years put actual wins on the board for celebrities going after deepfakes.
Megan Thee Stallion won the first jury trial. Florida federal court, 2025. A blogger who had harassed her for years promoted a fabricated sexualized video of her. The jury awarded $75,000, later adjusted to about $59,000 on procedural grounds. The part that stings more is that the defendant has to cover over $1 million of her attorney's fees. The core claims were defamation and harassment, but "promoting a deepfake" stood as its own item of damages for the first time.
Bang Chan of Stray Kids pried open cross-border anonymity. In September 2025, a California federal court granted his request forcing X to hand over a deepfake poster's registration email, phone number, and IP address. Only with the identity in hand could his civil suit in Seoul move forward. That route, discovery in America, lawsuit at home, is now the textbook play for artists' lawyers worldwide.
xAI is being sued in four jurisdictions at once. Ashley St. Clair, the mother of one of Musk's children, sued in New York over Grok generating pornographic deepfakes of her. In March, a class action was filed in California with three victims, one of them a minor. In June, the British MP Jess Asato filed in the High Court in London. Minnesota's attorney general sued under the state's nudification law, which carries up to $500,000 per violation, and xAI countersued, arguing the law is unconstitutional. SpaceX's public offering documents set aside more than $500 million for Grok-related litigation.
Johnny Cash's estate sued Coca-Cola. November 2025, over an NCAA ad featuring a voice that sounds a lot like Cash. It is the first major test of Tennessee's ELVIS Act, the first state law in the country to write "voice" into personality rights. The twist is that the defendant used a human sound-alike singer, no AI involved, so the court will have to answer whether the law covers flesh-and-blood imitation.
George Carlin's estate went after an AI podcast. In 2024, an "AI-generated Carlin comedy special" came down within a week of going up. The parties settled and the court issued a permanent injunction. The case is widely treated as the first template ruling for AI impersonation of a celebrity's voice.
India delivered arrests and injunctions in the same stretch. The Rashmika Mandanna case, the one that made "deepfake" a household word in India, ended with the main suspect arrested in January 2024. The motive was absurd in a typical way. He wanted more followers for a fan account he ran. Bollywood stars have since lined up at the Delhi High Court for personality rights injunctions, from Amitabh Bachchan to Aishwarya Rai.
One thing keeps getting written up wrong and should be corrected on Scarlett Johansson's behalf. She never sued OpenAI. In the 2024 "Sky" voice episode, ChatGPT's voice sounded so much like her that her friends could not tell the difference, and she had turned down the voice job earlier. Her tool was lawyers' letters. OpenAI pulled the voice, and later documents showed a different actress had recorded it. What she did do was point the episode's entire publicity at one goal, a public call for federal legislation. That is the other half of how celebrities fight back. Win the narrative, push the law.
The legal puzzle, what a celebrity can actually use in 2026
What can a deepfaked celebrity reach for in 2026? Ranked by how usable each one is.
-
The TAKE IT DOWN Act, in force, the only fully effective federal law. Publishing nonconsensual intimate imagery is a federal crime, and platforms must remove it within 48 hours. It protects any "identifiable individual," stars and ordinary people alike.
-
The right of publicity, the most usable civil claim right now, which is the mainstream view among legal scholars. But it is a state-law patchwork. Tennessee's ELVIS Act is the most aggressive, New York and California have their own versions, and there is no unified federal right.
-
The DEFIANCE Act, pending. Civil damages starting at $150,000. It has passed the Senate unanimously twice and is still stuck in the House.
-
The NO FAKES Act, pending. It would create a federal property right in everyone's face and voice, with statutory damages up to $750,000 per work and protection lasting 70 years after death. It cleared the Senate Judiciary Committee unanimously in June 2026, with RIAA, SAG-AFTRA, Google, and OpenAI all publicly supporting it. It has yet to get a floor vote.
-
The overseas toolbox. The EU requires deepfake labeling from August 2, with fines up to 15 million euros. In the UK, creating a sexual deepfake became a crime in February, and even requesting one is illegal. India's new rules require AI content labels to cover at least 10% of the frame, and nudity-related deepfakes must come down as fast as 2 hours after a report. How Korea rewrote its laws and raised its sentences is covered in our K-pop piece. Denmark is pushing a bill that would give every person a "copyright" in their own likeness.
The full legal tracker, state by state, country by country, updated monthly, lives on our Deepfake Law Tracker page, so I will not unpack it all here.
Celebrities also have one layer of equipment ordinary people do not, a protection industry. CAA built a "digital asset vault" for its artists. WME partnered with the detection firm Loti, which reports a 95% takedown success rate at an average of 17 hours. SAG-AFTRA has written digital replicas into every contract from film and TV to games to advertising. Even Fortnite drew a union grievance for using a dead actor's AI voice. Before the law showed up, Hollywood built its own protection first.
Consent is the only dividing line
Here I have to add a piece that many articles like this one leave out. Deepfake technology is not evil in itself.
After throat cancer took Val Kilmer's voice, AI restored it. David Beckham delivered an anti-malaria appeal in nine languages he does not speak. In Paul McCartney's "Find My Way" music video, a VFX team turned the 82-year-old back into his Beatles-era self. The singer Sevdaliza got permission from Grimes and Madonna for the face swaps in her video, and the director said the idea was to take a technology that usually gets weaponized against artists and use it for a collaboration instead.
Same technology. The dividing line is one question. Did the person behind that face say yes?
That is also the line lawmakers around the world are drawing. NO FAKES has carve-outs for news, documentary, commentary, and parody. The EU has its labeling duty. SAG-AFTRA has its consent clauses. All of it is built around consent plus disclosure. The industry has even grown a job few people talk about, the "deepfake actor," an unknown performer who does the acting while a famous face goes on top. Their labor rights are the setup for the next contract war.
When even the real can be called fake
One last harm, quieter than the rest. Once deepfakes get common enough, the damage turns around. Real evidence can be called fake too.
This is not a thought experiment. In 2023, Tesla's lawyers argued in court that a 2016 video of Musk speaking publicly about Autopilot safety "might be a deepfake," trying to get the evidence thrown out. The legal scholars Chesney and Citron had already named this phenomenon the "liar's dividend," and this was the first time a celebrity-level party carried it into a courtroom. Research has also found that when public figures dismiss a real scandal as a deepfake, their support can actually go up.
Follow this line a little further, and here I am extrapolating on my own. A celebrity's real scandal tape can be washed clean with one sentence, "it's AI." The same sentence can deny a police encounter filmed on an ordinary person's phone. At that point, what gets lost is much bigger than any one person's reputation. Seeing is believing, the rule society has always run on by default, stops working. The tech world's bet on a fix is content credentials, C2PA, which would give every photo and video a verifiable cryptographic signature at the moment of capture. Whether that spreads may matter more over the next five years than any single deepfake law.
FAQ
Are celebrity deepfakes illegal? It depends on the content. Porn deepfakes, publishing them is a federal crime, punishable by up to 2 to 3 years under the TAKE IT DOWN Act, and 46 states have laws of their own on top. Scam ads fall under federal wire fraud, and a pending bill would raise sentences for AI fraud to 30 years. Parody and commentary are protected by the First Amendment. For how the rules differ country by country, see our law tracker page.
Can celebrities actually win in court? Since 2025, yes. Megan Thee Stallion got a jury verdict and a seven-figure fee shift. The Carlin estate got a settlement and a permanent injunction. Bang Chan got a cross-border discovery order. xAI is being sued in four jurisdictions. The claims are all off the shelf, defamation, right of publicity, state deepfake laws, and the NO FAKES federal right on its way.
Who gets deepfaked the most? Measured by scam abuse, Musk, at 24% of celebrity scam deepfakes. Measured by search demand, research points to the streamer Pokimane. The structure matters more than any ranking, though. The top ten by search volume are all women, and women are targeted 4.5 times as often as men. We do not run victim leaderboards.
How do I spot a celebrity deepfake scam ad? Three red flags. First, a celebrity pitches you an investment, a supplement, or a giveaway "personally." Real endorsements run through the brand's official channels. Second, you have to pay a small fee upfront to claim a big prize, shipping and account fees included. Third, manufactured urgency, "today only." For hands-on verification, reverse image search and checking official channels, see our deepfake spotting guide.
What should I do when a celebrity deepfake shows up in my feed? Do not share it. That includes "sharing to debunk," which still spreads it. Use the platform's reporting channel to flag AI or impersonation content. If it is sexual, platforms are legally required to remove it within 48 hours, and if they do not, you can file with the FTC (TakeItDown.ftc.gov). If you or someone you know is a victim, the full action checklist is on our law tracker page.
Are funny celebrity face swaps and parody videos legal? Parody is protected speech, but the edges keep getting sharper. Nothing sexual, that is illegal. Nothing used to sell products, that infringes the right of publicity. Inside the EU, AI labeling is mandatory from August. A practical safety line, make it obvious at a glance that the video is fake, and do not use it to sell anything.
Closing
Celebrities have been the crash test dummies of this collision. The damage hits them first, the data gets collected from them, and the protections get shaped on them before anyone else. Taylor Swift's 17 hours measured how fast platforms react. Grok's 11 days measured how fast generation runs. Every celebrity lawsuit and every bill named after a victim over the past two years has been working on the same question. When anyone's face can be used by anyone, who owns "you"?
The answer is taking shape. Consent is becoming the dividing line, platforms are picking up hard deadlines, and faces and voices are turning into legally protected property. Celebrities got these protections first, but ordinary people getting them is only a matter of time. DEFIANCE and TAKE IT DOWN both write their protections for the "identifiable individual." Swift qualifies. So do you.
Next time a shocking video of some star scrolls past you, hold off on believing it, and hold off on sharing it. That face may not belong to that person.
Sources
- Security Hero, State of Deepfakes 2023 (98% / 99% / 94% figures)
- Resemble AI quarterly deepfake incident reports (2025 Q1/Q3)
- FBI IC3 2025 annual report; FTC impersonation fraud data (2025); BBB Scam Tracker
- CCDH monitoring report on Grok sexualized imagery (2026-01)
- Court and congressional records, Congress.gov (DEFIANCE / NO FAKES / TAKE IT DOWN), Megan Thee Stallion case, Bang Chan discovery order, Cash v. Coca-Cola, Carlin settlement
- WIPO Magazine, AI deepfakes in the entertainment industry (2022)
- Arnwine, When Deepfakes Make Celebrities a Dime a Dozen, UF Journal of Technology Law & Policy (2025)
- McAfee deepfake deception list (2025); CreateUGC search study (2026); Kyodo / Japan Today (Japan's 43,000-item count)
- Variety / Rolling Stone / Hollywood Reporter / NPR / Engadget / Fortune (case reporting)