Evidence Pack: How Digital Forensics Caught the AI Deepfakes Targeting NYC's Mayor

In early 2026, following the unsealing of a new batch of Justice Department documents related to the late financier Jeffrey Epstein, a series of explosive images flooded social media platforms. The visuals appeared to provide visual confirmation of a sprawling conspiracy, weaponizing the document release to target a sitting politician.[1]

The photos purported to show New York City Mayor Zohran Mamdani as a young child, alongside his mother, the acclaimed filmmaker Mira Nair. In the images, the pair were seen posing intimately with Epstein, his associate Ghislaine Maxwell, and a rotating cast of high-profile figures including former President Bill Clinton, Bill Gates, and Jeff Bezos.[1][2]

The images racked up millions of views across platforms like X, Facebook, and Truth Social within hours. The viral spread sparked immediate outrage, fueling a wave of online conspiracy theories that ultimately culminated in physical protests outside Gracie Mansion, the official residence of the New York City mayor.

But the photographs were entirely fabricated. A coalition of digital forensics experts, news organizations, and independent fact-checkers quickly dismantled the narrative, proving beyond a doubt that the images were AI-generated deepfakes designed to deceive the public.[2][3]

This evidence pack breaks down exactly how investigators debunked the viral smear campaign. By examining the digital metadata, chronological logic, and source provenance, researchers provided a masterclass in how to catch synthetic media in the wild.

The most definitive proof of fabrication came from the images' underlying digital architecture. As AI image generators have grown more advanced, tech companies have begun embedding imperceptible signals into the files they produce to ensure they can be identified later.

Investigators from BOOM Live and CBS News ran the viral photos through Google's SynthID detector. SynthID is a specialized digital watermarking tool designed to identify content that has been created or significantly altered using Google's proprietary artificial intelligence models.[1]

The SynthID scan returned a positive match across the entire batch of viral images. This cryptographic evidence confirmed that the pixels had not been captured by a camera lens, but rather generated by a machine learning algorithm.[1]

Beyond hidden digital watermarks, fact-checkers identified glaring logical errors in the images' timelines. AI models often struggle with chronological consistency, and this batch of deepfakes was no exception.[2]

The photos consistently depicted Mamdani as a toddler or a very young boy. However, the purported context for the images—the minimal factual anchor used to sell the lie—was a 2009 film screening after-party mentioned in a single email within the newly released DOJ files.[2]

In 2009, Mamdani was 18 years old, not a toddler. Furthermore, forensic analysts at AFP noted that across the series of fake photos, the child appears to age rapidly from a baby to a preteen, while the adults around him—including Epstein and Clinton—do not age a single day, remaining frozen in time.[2]

The final nail in the coffin was the provenance of the images. Open-source intelligence (OSINT) researchers tracked the photos back through the social media ecosystem to find their absolute origin point on the internet.[3]

The European Broadcasting Union and other disinformation monitors traced the initial uploads to a self-described parody account on X. The account's own biography openly stated that its purpose was to create "high-quality AI videos and memes," effectively admitting to the fabrication from the start.[3]

Despite the account's satirical bio, bad actors quickly stripped the images of their original context. They downloaded the files and re-uploaded them to alternative platforms like Truth Social as authentic historical photographs, deliberately bypassing the parody label.

Open Measures, a threat intelligence firm, documented how conspiracy theorists used the minimal context of Nair's name appearing in the 2009 DOJ email to craft a viral, visually supported lie. The AI images served as the emotional hook to make the baseless narrative feel tangible.

The real-world consequences of this digital fabrication were immediate and tangible. The misinformation successfully bridged the gap from the internet to the physical world, driving angry citizens to protest outside the mayoral residence based entirely on synthetic evidence.

Mayor Mamdani addressed the deepfakes publicly, describing the experience of seeing "patently photoshopped" images of his family spread globally as "incredibly difficult." He noted the old adage that a lie can travel halfway around the world while the truth is still putting on its shoes.

He used the incident to call for urgent regulatory frameworks governing synthetic media, warning that current civic and legal systems are "ill-equipped for the speed and the reach of the technologies in front of us."

Ultimately, the rapid debunking of the Mamdani deepfakes serves as a successful case study in modern digital literacy. By combining cryptographic watermarks, basic chronological logic, and source tracing, fact-checkers were able to definitively neutralize a sophisticated piece of synthetic propaganda, offering a playbook for future digital investigations.