A fake AI image of actress Mia Goth, posted by a "verified" Instagram fan account, collected nearly 500,000 likes, spread through celebrity gossip sites and was even summarised by a Google AI overview as fact. The case, reported on October 15, 2026 by The Verge, is a clean example of how AI-generated likenesses move from a social post to search results, and of how little platform rules require in return.
This post summarises what The Verge found, compares the rules at Meta and YouTube as the reporting describes them, and lists practical ways to check an image before you share it. It is not a claim that any particular celebrity said or did anything; the point of the case is that the image was fake.
TL;DR
| Question | Answer (source: The Verge unless noted) |
|---|---|
| What was the fake? | An AI image of Mia Goth in a T-shirt reading "MY EX IS MY BIGGEST FAN", set to a Beyoncé song, days after news of her divorce filing |
| Where? | Instagram account @miagoth_, over 1 million followers, blue check, labelled a fan account |
| Reach | Nearly 500,000 likes, including from people in the entertainment industry |
| Spread | Gossip outlets, then Google search, where an AI overview presented it as a swipe at her ex |
| Outcome | Instagram removed the account for impersonating a public figure after The Verge asked Meta about eight accounts |
| Labelling | Meta exempts still images from its AI-label requirement |
| Alternative | YouTube's likeness detection, now open to every adult account |
How the Goth fake spread
The post looked like a celebrity statement. Goth, the report notes, has little or no social media presence, which should have raised doubts. But the account was verified, had more than a million followers, and carried the label of a fan account. Meta lets users buy blue checkmarks, even if the account is not run by a prominent figure or an official page.
Goth's representative Christine Tripicchio told The Verge that the fact that the account qualified for a blue check "is disconcerting on a fundamental level" and that it is "incomprehensible that Instagram evidently prioritizes the dissemination of forged and manipulated images over the protection and integrity of a public figure's likeness." The representative said they had complained to Instagram before about the account posting "counterfeit, AI images."
Because of the divorce news, gossip blogs ran the post. Their coverage then fed Google: searches for the actress and the shirt triggered an AI-generated overview that said she was taking "a clear swipe at her ex," and a results section titled around the shirt appeared prominently. That is the loop to notice: a fake post becomes a source for blogs, and the blogs become sources for an AI summary, which makes the fake look corroborated.
A wider pattern: "AI slop" featuring real people
The report places the case in a larger flow of content. Examples it gives include:
- Videos of an AI Taylor Swift lip-syncing and rapping, with 1.8 million views, and similar clips of fictional versions of other famous people, with views in the millions.
- Fake wedding and movie-premiere videos that follow set formats.
- Accounts that post sexualised AI imagery of real women, some of which route to paid subscriptions. We are not describing those accounts further here.
- A profile advertising $29-per-month courses on making money from AI content.
The Verge says Meta removed some of eight accounts it asked about, including the Goth one and a Scarlett Johansson one, while another account with over 300,000 followers still posts a mix of authentic and apparently AI-generated images of Johansson with no AI disclosure, in line with Meta's rules. A Meta spokesperson, Alex Dziedzan, said: "We do not permit accounts to use someone's likeness to pose as that individual, regardless of whether the content is authentic or AI-generated."
What the platform rules actually require
| Platform | What the reporting says |
|---|---|
| Meta (Instagram, Facebook) | Labels required for some content such as realistic videos and AI voices; still images are exempt. An "AI-generated profile" label applies to whole accounts whose avatar does not exist in the real world. Impersonation of a real person is prohibited, enforced after reports |
| YouTube | Likeness detection scans for visual matches, recently also sound-alikes; people request takedowns; reviewers weigh public interest, newsworthiness and consent. It now covers every adult with a YouTube account. YouTube says it removed over 95 percent of videos reported through the program in the first half of 2026 |
| US law | The Take It Down Act (2025) criminalises distributing nonconsensual intimate imagery, including AI-generated content, and requires platforms to remove it within 48 hours. There is no national right of publicity law for other fake likenesses |
YouTube's approach has been described by other outlets, for example The Decoder on the expansion to all adult creators. Reported details include identity verification with a selfie and ID, setup in YouTube Studio, and takedown criteria that consider whether content is realistic, labelled as AI and uniquely identifies a person, with carve-outs for parody and satire. The Verge adds that the feature covered facial likeness, with sound-alikes added more recently. A system like this is opt-in and reactive: it helps the person who enrolls, and not the viewer who sees the fake.
Why labels on still images matter
Meta's exemption for still images is the detail that stands out. A still is the cheapest AI output to make and the easiest to share, and a caption or screenshot travels without metadata. If a platform labels only video and audio, the most common fake format arrives unlabelled. Watermarking and provenance can help where the generator cooperates; see our guide to Google's public SynthID website for checking media and the explainer on spymarks versus watermarks. Neither helps if the generator does not embed anything, and open-weight image models often do not.
The legal picture is unsettled too. The Verge notes that in the absence of national right of publicity laws, AI content featuring real people is "being litigated on the platform level." Courts abroad are moving; for example, we covered China's top court on AI clones and personality rights, and in the entertainment industry the question of consent has surfaced in stories such as Nicolas Cage refusing an AI waiver.
Why takedowns do not solve it
The report describes a whack-a-mole dynamic. Even when one page is taken down, others continue to push out similar content, and the recommendation algorithm "seems to like" the cottage industry of accounts that pump out AI imagery of real people. Removing the Goth account after a journalist's inquiry fixed one case, but it did not change the incentives: a verified-looking account, an emotional hook and a timely news peg are enough for reach in the hundreds of thousands.
Three structural gaps follow from the reporting:
- Verification is a product, not a credential. When a blue check can be bought, it stops signalling identity, yet readers still treat it that way.
- Reporting is reactive. The Goth representatives had complained before; action came after press attention.
- Search amplifies the loop. Once gossip sites repeat a fake, an AI overview can present it as consensus, so a platform problem becomes a search problem.
For public figures the cost is reputational and legal: the fake put a pointed message about a divorce into her mouth, which the report calls a potential liability for the parties involved. For everyone else, the lesson is that the same tools and incentives apply to ordinary people, which is why YouTube's decision to widen likeness detection to every adult account matters more than celebrity cases alone.
How to check a celebrity image before sharing
- Look at the account, not just the image. A fan page with a purchased check is not an official source. Check whether the person has an official account and whether the post appears there.
- Search for the same claim in reputable outlets. If only gossip aggregators carry it, treat it as unconfirmed. Remember that aggregators can copy each other and amplify a fake.
- Distrust AI overviews on breaking celebrity stories. An overview that cites gossip blogs is only as good as those blogs.
- Check provenance tools. Use Google's SynthID website for images from models that embed it, and note that a negative result does not prove an image is real.
- Zoom in. Hands, text on clothing, jewellery and background details are still common failure points, though the best generators are improving.
- Report impersonation. Instagram removed the Goth account for impersonating a public figure after it was flagged to Meta.
- If you are a creator, enrol in likeness detection where it is offered and keep records of your official accounts.
For readers concerned about fraud rather than gossip, deepfakes are also used in scams; see our posts on Google's on-device scam detection in Singapore and on AI agents that bait scammers. The Video Turing Test explainer covers how close generated video has come to passing as real.
What to watch
- Whether Meta extends labels to still images or changes its rules on purchased verification for accounts that post likenesses.
- Whether search providers change how AI overviews treat unverified celebrity claims.
- Whether the US adopts a federal right of publicity or extends Take It Down-style duties beyond intimate imagery.
- How well opt-in likeness systems scale when every adult can enrol.
Related reading
-
Fake AI images of a punched protester at a Trump rally: what SynthID showed
-
Google's SynthID website: check whether media is AI-generated
Details are accurate as of October 15, 2026, from The Verge's reporting; platform policies change.