Nobody is safe from deepfake AI manipulation, even celebrities
Instagram posts should have raised more red flags in the past. This image apparently shows actress Mia Goth taking a photo in a T-shirt reading "My ex is my biggest fan," just days after she filed for divorce from actor and filmmaker Shia LaBeouf. To really drive the point home, the song "Single […]
In the past, posting on Instagram should have raised more red flags. The image purportedly shows actress Mia Guth posing in a t-shirt with the message: My ex is my biggest fan, just days after filing for divorce from actor and filmmaker Shia LaBeouf. To really put this into perspective, Beyoncé's song "Single Ladies (Put English In It)" was added to this post.
It was an impossible character for the pearl star, who is not very active on social media. But the Instagram account that posted it, @miagoth_, had more than a million followers and was marked with a blue "verified" checkmark. Within days, the image garnered nearly 500,000 likes, including from peers in the entertainment industry, despite the page being tagged as a fan account.
As The Verge reported, the image was created by AI and was posted by a user account that Goth's representative said has previously sued Goth on Instagram for posting "fake and AI images." Meta allows users to purchase blue "verified" checkmarks even if the account is not run by a prominent figure or is not an official page.
"The fact that this fan account meets Instagram's criteria to qualify for a 'blue check' is troubling on a fundamental level," Goth representative Christian Tripicchio told The Verge in an email. Beyond that, it is incomprehensible that Instagram would openly prioritize the publication of fake and manipulated images over the protection and integrity of a public figure.
Due to the news of her divorce, this post was published in popular gossip media, and this coverage in turn published this post through a Google search. Searches for the actress and the shirt yielded an AI-generated overview of Guth, in turn, citing the aforementioned blogs as "a jab at his ex Shia LaBeouf following their recent divorce filing." A section titled "Mia Goth Turns Heads on Instagram Wearing a Cheeky 'My Ex Is My Biggest Fan' T-Shirt" is featured prominently in Google search results, along with links to social media posts discussing the image. Instagram eventually removed the account for impersonating a public figure.
Since consumer AI tools became available several years ago, some of the most memorable — and controversial — use cases have involved celebrities. The human subjects in question have historically not reacted well to being digitally cloned and used as toys: a Drake sound-alike created by artificial intelligence was removed from YouTube following copyright strikes. Scarlett Johansson has hired a legal team after OpenAI released a voice for ChatGPT that sounded like her without her consent. Ariana Grande has reportedly suspended an account that was posting her AI-generated content last year. Taylor Swift has publicly endorsed Kamala Harris for president in 2024, saying she encouraged her to clarify her position after Donald Trump posted fake pictures showing her endorsing him.
It was a moment when everyone, even the rich and famous, struggled with the norms surrounding a new technology. Now, the level of AI content online is on the rise – and it's unclear whether anyone will try or be able to stop it.
AI content depicting female public figures is often overtly sexual. The profile, which exclusively publishes AI-generated images of Johansson — often featuring bare feet prominently — includes contact information so companies can advertise on her account, along with links to Patreon and Telegram channels that sell subscriptions. Off-platform content used Johansson's name and image (AI) to promote adult content. Another account posts real AI-generated images of actors and musicians like Millie Bobby Brown, Zendaya, and Grande, who are always half-naked or in sexual situations. Some accounts have their own niche, such as "Strange Things Actors Who Marry Each Other".
After The Verge asked Meta about the eight accounts that posted celebrity deepfake content, spokesperson Alex Dezidzan said the company had removed some of the accounts. Two of them - Goth and Johansson - seem to have been removed.
"We do not allow accounts to use a person's likeness to impersonate that person, regardless of whether the content is authentic or generated by artificial intelligence," Dziedzan said in an email.
Taylor Swift's got a drink in her hand, she's rapping along to "Burning Bridges," rapping and all. Wearing a dress from the fateful 2009 VMAs, she sings, "I'm out of [Quaalude]/It's tingling my body." The moment is clear that the video is AI. It has 1.8 million views on Instagram.
One person commented: “I don't think her PR team would like her saying the N-word.
Instagram is full of videos like this, which often inexplicably follow very specific formats. A clip of Shrek's AI doing the same thing as Swift's AI has 3.1 million views. Here's Steve Jobs belting out a different Drake song with a bottle of champagne (371,000 views). Also, a similar version of Lionel Messi (2.1 million views) was posted by another account. Here's Formula 1 drivers Lewis Hamilton and Max Verstappen lip-syncing to "APT" by Bruno Mars and Rose (2.9 million views). Another clip shows Swift's AI wedding to Travis Kelce's AI at Madison Square Garden.
"Yes, she was wearing a white dress," one fan checked with the help of Truth.
In the absence of national advertising rights laws, AI-generated content featuring real people is objectionable at the platform level. (The US Takedown Act, passed in 2025, specifically criminalizes the distribution of non-consensual confidential images, including AI-generated content, and forces platforms to remove it within 48 hours.)
YouTube has similarity detection, perhaps the most powerful deepfake monitoring and removal system. Under the program, YouTube scans its platform for visual matches to a person — and recently similar sounds — and allows them to request that the content be removed. Not every takedown request will be accepted. The company says it considers public interest, newsworthiness and satisfaction when evaluating takedowns. The program was initially open to the general public such as celebrities, content creators, politicians and journalists, but now covers any adult with a YouTube account. In the past, the company has said that the number of takedown requests it receives is relatively low. Creators VP of product Amjad Hanif said at a press conference in March that creators want to be aware of what's posted on YouTube about them, but it often comes down to "very benign or additive to their overall business."
YouTube spokesman Jack Malone noted in an email to The Verge that the company's similarity detection was developed in collaboration with the entertainment industry. Malone said that in the first half of 2026, YouTube removed more than 95 percent of videos reported through the program for removal.
AI-generated content on meta platforms often has tags identifying it as such. Some content, such as real videos of people or audio generated by artificial intelligence, must be tagged with artificial intelligence. Static images, on the other hand, are exempt from the tagging requirement – meaning that Mia Goth's fake image does not need to be disclosed under the meta rules. The company also introduced "AI-generated profile" tags that apply to an entire account where the avatar depicted doesn't exist in the real world after an influx of AI influencers took the platform by storm.
But there's a cottage industry of accounts posting AI-generated images of real people that the algorithm seems to like. Some content isn't overtly evil, just kind of weird: there's a genre on Facebook, for example, of AI-generated videos of the first movie where the main actors (in costume) walk the red carpet alongside the cast for the remake. A profile advertises $29/month online courses on how to make money with AI content.
AI-generated celebrity content also has the capacity to actually trick people into thinking it's real, such as the image (and account) featuring Goth. One particular post that went viral shows the actor mocking LaBeouf, who has been accused of harassment by another ex-partner, and was filmed in 2015, saying he would "kill [Goth]" if he kept arguing with her. The fake image was attention-grabbing, obscure, and timely—in other words, great for social media virality but a potential liability for the parties involved. This incident demonstrates the failure of safety and trust regulations that are supposed to protect against this type of confusion.
On social media, AI content is freely mixed with standard reposts of real images. And even when a page is deleted, there are others that continue to post content: Another Instagram account — still alive — with more than 300,000 followers posts a steady stream of photos and videos of Johansson that appear to be a mix of authentic images and AI-generated content. One of the most popular posts is what appears to be an AI image of Johansson with nearly half a million likes. There are no AI disclosures, per the meta rules. Deepfake Johansson is lying on a couch and her bare legs are prominently displayed. Obviously he's not - but his fans in the comments either don't know or don't care.