A Fake Mia Goth Photo Hit 500,000 Likes Before Instagram Acted

Instagram removed the account @miagoth_ for impersonating actress Mia Goth after an AI-made image of her collected nearly 500,000 likes in a matter of days. It was fake. Industry peers liked it anyway.

The picture claimed to show Goth days after her divorce filing from actor and filmmaker Shia LaBeouf, wearing a T-shirt reading "MY EX IS MY BIGGEST FAN" The Verge. The post paired the photo with Beyoncé's song "Single Ladies (Put a Ring on It)". The filing ended a marriage of nearly 10 years. Goth and LaBeouf married on Oct. 10, 2016 and separated in July 2025, according to divorce records cited by the Los Angeles Times.

The account used the handle @miagoth_, held more than one million followers, and carried a blue verified checkmark while also labeled as a fan account. Meta allows users to purchase a blue check even if the account is not run by a prominent figure or is not an official page. That mix of large audience, trust signal, and small disclosure label traveled quickly. Likes from entertainment industry peers helped push the image further before its synthetic origin, meaning it was made by AI, was widely understood.

Goth's team said it had seen this account before. Representative Christine Tripicchio said they had previously complained to Instagram about the account posting "counterfeit, AI images" of Goth. Instagram acted only after the T-shirt image went viral.

The image then appeared in search. Google searches for Goth and the shirt triggered an AI-generated overview, an auto-written summary at the top of results, saying Goth was taking a swipe at her ex following the divorce filing. A synthetic post on one platform became a synthetic summary on another, with no check in between.

Similar cases have involved other stars. Donald Trump posted fake images suggesting Taylor Swift endorsed him. Swift later publicly endorsed Kamala Harris for president in 2024, saying she was moved to clarify her position after the fake endorsement images. Scarlett Johansson hired a legal team after OpenAI released a ChatGPT voice that sounded like her without her consent. An AI-generated Drake sound-alike song was wiped from YouTube following copyright strikes. Ariana Grande is said to have blocked an account posting AI-generated content of her.

Some defenses exist. Meta's Authentic Identity Representation policy covers using a person's image(s), name, or likeness with the aim to deceive others, per Meta. The company said in October 2024 that nearly 500,000 public figures were being protected from having their likeness misused in scams, using tools including face-matching for scam ads. Swift filed to trademark her voice and likeness to ward off AI deepfakes, with applications covering two audio clips and one image of herself filed on Friday, April 24, according to Reuters. X previously blocked users from searching for Taylor Swift after fake sexually explicit images of her spread.

The broader context here is an enforcement gap, not a detection gap. Synthetic celebrity images are now cheap to produce, verified distribution is cheap to rent, and peer engagement provides rapid social proof. Policy language on impersonation exists. Face-matching for scam ads exists. Trademark filings exist. None of them intervened before half a million likes accumulated on a fan-labeled, checkmarked account that the subject's representative had already flagged.

In my view, two design choices deserve scrutiny from people who build these systems. First, paid verification still borrows authority from an older idea of identity proof, while functioning as a subscription badge. Users read the check as authentication. The fan-account label did not counteract that reading here. Second, AI summaries in search currently inherit the credibility of the index without inheriting its provenance controls, meaning they do not show where each claim came from. When the index contains viral slop, the summary repeats it in an authoritative voice.

Looking at what this means for teams working on trust and safety, the Goth case suggests likeness enforcement cannot sit only in post-hoc takedowns, or removals after harm is done. Pre-viral signals were present: a prior complaint, a fan-account label paired with a celebrity handle, and rapid velocity on a sensitive personal event. I have watched my own kids take a viral post at face value before asking who posted it. The optimistic read is that all three are machine-readable. The systems that monetize velocity could also be tuned to pause it when identity, synthetic media risk, and real-world harm intersect.