Epstein Files Disinformation: How Fabricated Celebrity Claims Hijacked the Truth

On February 20, 2026, The People's Voice — the rebranded incarnation of NewsPunch and before that YourNewsWire, founded by Sean Adl-Tabatabai — published the headline: "Epstein Files Reveal Leonardo DiCaprio Ate 70 Pounds of Child Meat on 'Cannibal Diet.'" The claim alleged DiCaprio discussed a "cannibal diet" in emails with filmmaker Woody Allen. ^[1]

What the files actually contain regarding DiCaprio: a "Django Unchained" teaser trailer premiere reference, an invitation for a private film screening, and a 2009 email in which Epstein asked UK politician Peter Mandelson whether anyone might want DiCaprio's endorsement for a non-U.S. sponsorship deal. ^{[1] [30]} DiCaprio authored none of the documents and had no correspondence with Epstein in the released materials. ^[1] Media Bias/Fact Check rates The People's Voice as having "zero credibility" and says it has more false claims flagged by IFCN-accredited fact-checkers than any other website. ^[9]

The story bounced from The People's Voice to X, then to international tabloids in Pakistan, India, and Nigeria before Snopes published its debunk on February 25, 2026. The cannibalism framing was deliberately extreme — engineered to trigger disgust-based sharing. ^{[1] [11]}

On February 11, 2026, The People's Voice published an article and video claiming the Epstein files exposed Ellen DeGeneres as a "prolific cannibal," based on an alleged "whistleblower" audio recording. Four independent analysts told PolitiFact the audio was almost certainly AI-generated. Snopes and PolitiFact both independently rated the claim FALSE. ^[20]

DeGeneres appears in the actual files only in news clippings and feature-story excerpts that mention her name in passing. No cannibalism references in the files mention her. ^[20] This establishes a clear production pattern: The People's Voice ran at least two identical operations (DiCaprio and DeGeneres) within nine days, using the same structural lie — grafting a celebrity name onto real but unverified cannibalism references that appear elsewhere in the files. ^[20]

Vietnamese spam operations running Facebook pages fabricated a story that Stephen Colbert and Jon Stewart secretly hosted an independent show called "Freedom and Justice" about Giuffre's death, with its first episode reportedly reaching 1 billion views. ^[3] A variant claimed Jon Stewart alone livestreamed about Giuffre from his home, garnering 3.2 billion views; Meta's transparency data confirmed the "Afternoon Daily" Facebook page pushing that version was managed from Vietnam. ^[2]

Virginia Giuffre died by suicide on April 25, 2025, in Neergabby, Western Australia. ^[15] The fabricators exploited her death by constructing a narrative of posthumous vindication. The view count figures exceed the viewership of any media event in history and exist solely to suggest suppressed viral magnitude. Content linked to northwavenews.com, an ad-farm site in the Vietnamese spam network ecosystem. Lead Stories confirmed the AI-image patterns. ^[12]

An X account operated by "Paul White Gold Eagle" — who markets himself as a spiritual practitioner, not a journalist — posted the claim that Michael Feldman's company FGS Global was named as an Epstein co-conspirator in the DOJ files. The claim was amplified during the same period Savannah Guthrie's mother was missing, creating an implied narrative that the journalist was being silenced for her connection to Epstein. ^{[4] [5]}

FGS Global is not mentioned in the Epstein files. The Glover Park Group — a predecessor firm that Feldman helped found in 2001, which merged into FGS Global in 2021 (two years after Epstein died) — appears twice in the documents, incidentally, with no implication of conspiracy. This hoax layered two real events to manufacture a false causal link — a hallmark of sophisticated disinformation. ^[5]

British artist Alison Jackson — whose work is held in the Parliamentary Art Collection, the National Portrait Gallery London, SFMOMA, and other public collections including the Pompidou Centre and Fotografiska ^[32] — posted a staged lookalike photograph to her own verified accounts on November 17, 2025, showing Trump and Epstein lookalikes receiving a foot bath from a young woman, credited under her name and artist hashtags. ^[6]

After the DOJ's January 30 release, users stripped the image of its context and redistributed it with claims it was authentic evidence from the Epstein files. Because Jackson's images are intentionally hyperrealistic, they are uniquely vulnerable to bad-faith misuse. Snopes catalogued at least 8 distinct fabricated images circulating as Epstein file content in February 2026, including AI-generated images of Chuck Schumer, Hakeem Jeffries, and Nigel Farage, plus an AI-animated video that inserted minors into an authentic 1997 Trump/Epstein party photo. ^[7] A separate fabricated image showing Epstein alongside Bill and Hillary Clinton and other public figures carried a SynthID watermark — Google's AI-content identifier — confirming synthetic generation. ^[29]

A fabricated quote attributed to Attorney General Pam Bondi — "If we prosecute everyone in the Epstein files the whole system collapses!" — spread across Facebook, X, Instagram, and Threads in February 2026. The DOJ confirmed via email to Snopes that "the quote is fake and has been completely fabricated." No recording, transcript, or credible media report of Bondi making this statement exists. The quote was designed to suggest the current administration was suppressing prosecution — weaponizing public skepticism about the DOJ's handling of the release. ^[21]

The most significant previously underdocumented mechanism of image fabrication was X's own AI tool, Grok. Bellingcat's Digital Hate and Harms team investigated 31 separate Grok requests between January 30 and February 5, 2026, to "unblur" or "unredact" images from the Epstein files. ^[19]

The results: 27 of 31 requests — 87% — produced fake AI-generated images. Images ranged from "believable to comically bad, such as a baby's face on a young girl's body." Some posts accumulated millions of views on X before removal. Content fabricated included: unblurred faces of redacted children and women; photos falsely showing political figures with Epstein; and fabricated photos of Mayor Zohran Mamdani as a child with Epstein. ^{[19] [13]}

After Bellingcat contacted X on February 6, guardrails partially activated: Grok declined 14 of 16 subsequent requests (Feb 7–9) — a compliance rate dropping from 87% to 12%. However, in the window before intervention, the content had already spread at scale. A separate Center for Countering Digital Hate analysis found Grok had publicly generated approximately 3 million sexualized images — including 23,000 depicting children — within 11 days, predating and contextualizing its behavior on Epstein imagery. ^[19]

By contrast, OpenAI's ChatGPT declined all equivalent requests, citing its policy against "sexualized depictions of minors or scenarios that imply sexual abuse." A France 24/AFP study found Grok Imagine produced "convincing fakes in seconds" for all five politicians tested. The detection signature for many fake images: they were shared cropped, but uncropped originals retained a Gemini logo — confirming Google's AI as a secondary generation tool. ^[22]

Beyond the individual hoaxes, a secondary disinformation layer operated in parallel: false claims about censorship, designed to prime audiences to distrust any correction or debunk they might encounter.

Users reported that typing "Epstein" in TikTok direct messages triggered automatic content moderation warnings and message blocking. This generated a large-scale censorship narrative, with California Governor Gavin Newsom calling for a state investigation. However, NPR and platform researchers found the blocking was inconsistent — some users could send the word, others could not — and TikTok attributed it to a data center power outage, not intentional censorship. ^[23] Eight academics who examined TikTok's content landscape after the U.S. ownership transition, publishing in Good Authority, found "no evidence of large-scale prevention of content about these topics." ^[31]

The meta-disinformation significance: false claims of censorship reinforced the conspiratorial frame that powerful forces were hiding information, amplifying credulity for fabricated claims about the actual file contents. Audiences primed to believe they were being censored were more likely to share unverified claims as acts of resistance.

The People's Voice (NewsPunch) used an identical structural technique for both the DiCaprio and DeGeneres hoaxes: the FBI interview summaries in the actual Epstein files do contain unverified references to cannibalistic ritual violence — these are unsubstantiated claims made to investigators, not proven findings. The People's Voice grafted celebrity names onto these real but unverified references to produce entirely fabricated specific claims. The structural lie is elegant: the underlying document reference is real; only the celebrity connection is invented. When fact-checkers point out the celebrity isn't mentioned, the response from true believers is: "Of course they redacted the name." The architecture is self-sealing. ^{[1] [9]}

The Epstein files disinformation story has a second, underreported layer: real harm caused by institutional failures entirely distinct from the external disinformation ecosystem.

Attorneys representing roughly 300 alleged victims called the DOJ's January 30 release "the single most egregious violation of victim privacy in one day in United States history." ^[18] The DOJ failed to redact 31 of 32 minor victims named in a single email in the release. ^[27] The department also published dozens of unredacted nude images of women whose ages could not be confirmed — some removed only after The New York Times flagged them. ^[28] Attorneys separately reported "thousands of redaction failures" affecting nearly 100 individual survivors. ^[17]

The DOJ subsequently confirmed it removed "several thousand documents" that mistakenly identified victims. The court-ordered certification by US Attorney Jay Clayton that no victim-identifying information had been published was demonstrably false at the time of release. ^[17]

These failures interacted toxically with the disinformation ecosystem. Real content failures — actual redaction errors in genuine FBI documents — made it structurally harder for audiences to dismiss claims that the files contained damaging unredacted material. The DOJ's credibility gap became a disinformation amplifier.

The Trump administration stated the January 30 release was effectively complete, while acknowledging reviewing approximately 6 million documents total but releasing only 3.5 million — meaning nearly half the reviewed documents were withheld or redacted without public explanation. This 2.5 million document gap provides ongoing material for new disinformation cycles: anything can be claimed to be in the unreleased portion. ^[24]

Criminal accountability remains essentially absent: beyond Epstein himself and Ghislaine Maxwell, no alleged co-conspirators have faced U.S. criminal conviction in connection with the Epstein trafficking case. This reality — spanning multiple administrations — is the structural engine beneath the conspiracy speculation. When official accountability fails, disinformation fills the vacuum. ^[24]

The most significant structural analysis of the Epstein disinformation wave comes from an Al Jazeera opinion piece by Yoav Litvin (February 26, 2026): the ecosystem functions as political distraction — even when users believe they are pursuing accountability. The mechanism: sensationalized fabrications (celebrity cannibalism, fabricated quotes) channel public outrage away from structural analysis toward emotional spectacle. Conspiracy narratives about "hidden enemies" displace demands for systemic reform with demands for individual prosecution. As Litvin writes, such conspiracy theories "transform systemic crisis and social instability into simple, emotionally gripping stories of social taboo-breaking, centred on hidden and untouchable enemies." ^[24]

The paradox is acute: the disinformation amplifies real public anger at real elite impunity, but redirects that anger in ways that produce no accountability and serve authoritarian political movements seeking a "strongman" narrative.

The Washington Post and NPR both documented that far-right and neo-Nazi groups were pre-positioned to exploit the Epstein file release for antisemitic recruitment. The 107% spike in ZOG (Zionist-occupied government) conspiracy mentions documented by Moonshot within 48 hours of the release was not organic public reaction — it was coordinated. ^{[8] [25]}

Neo-Nazi and white nationalist groups explicitly planned to use the Epstein revelations as a recruiting bridge, connecting elite child trafficking narratives to existing antisemitic frameworks about Jewish power. The celebrity cannibalism claims serve as emotionally engaging entry points into a broader extremist framework. The disinformation is not primarily about specific celebrities — it is a recruitment funnel. ^[25]

The amplification infrastructure behind these hoaxes was not organic. Vietnamese-based Facebook operations ran dozens of fake fan pages producing a constant stream of AI-generated celebrity clickbait for ad revenue — a pattern fact-checkers had documented and warned about months before the files' release. ^[12] X's own Grok chatbot actively manufactured fabricated imagery on request, and TikTok's recommender system amplified political content unevenly during a server disruption that fed censorship narratives of its own. ^{[19] [23]} Together, these platform dynamics let false content outrun fact-checks. ^{[12] [19]}

The January 30 DOJ release created a structurally perfect disinformation opportunity. Five converging factors explain why fake claims spread faster and farther than fact-checks:

Volume overload. 3.5 million pages, 180,000 images, 2,000+ videos. No human can read the release. Bad actors can claim anything is "in the files" with low risk of immediate contradiction — and, crucially, the DOJ's own admission that the release includes unverified public submissions means technically any claim could be defended as "in there somewhere." ^[16]

Pre-loaded conspiracy frameworks. Years of QAnon, Pizzagate, and elite-trafficking narratives had already convinced large audiences that celebrities commit child abuse. The Epstein case is the only instance where real criminal conduct at that level is documented — making it an irresistible anchor for conspiratorial elaboration. ^[14]

Emotional priming. The real content of the files is genuinely disturbing. Audiences are already in a state of moral outrage and heightened credulity when fake claims arrive.

AI fabrication speed. Grok produced convincing fake images in the same timeframe as the news cycle. France 24/AFP documented fabrication "in seconds." The tools to create convincing false evidence now operate at news speed. ^[22]

Platform structural failures. Meta's January 2025 rollback of American fact-checkers created the opening for Vietnamese spam networks to operate at scale. X's Grok actively fabricated evidence. These are not neutral platforms — their design choices are causal factors in disinformation spread. ^{[12] [19]}

The Epstein files disinformation wave of February 2026 is a case study in how real transparency events can be systematically weaponized. The same government action intended to bring accountability — 3.5 million pages of investigative records — paradoxically enabled the most concentrated disinformation wave in the case's history. The mechanism is straightforward: volume creates a verification vacuum that bad actors fill.

Three structural factors made this wave worse than it had to be. First, the DOJ's own decision to include unverified public submissions without labeling them created a structural legitimacy problem — fake content now carries the URL of the official DOJ website. Second, X deployed an AI tool (Grok) that actively fabricated evidence at scale, with guardrails only partially activated after a journalist investigation — not after platform detection. Third, Meta's rollback of American fact-checkers in early 2025 removed the primary friction mechanism against Vietnamese spam networks, which operated with essentially no resistance.

The hoaxes served multiple purposes simultaneously: they generated ad revenue for spam networks, provided celebrity-scandal content for tabloid amplification, and served as entry points for antisemitic extremist recruitment. None of these purposes require the hoaxes to be believed — they only require the hoaxes to spread. Spread is sufficient to accomplish the goal of distraction, division, and demoralization. ^{[24] [25]}

The corrective record is clear: every named fabrication in this report has been independently debunked by multiple accredited fact-checkers. Leonardo DiCaprio, Ellen DeGeneres, Jon Stewart, Savannah Guthrie's husband, and Pam Bondi have all been cleared of the specific claims made about them. The real Epstein files contain genuine, disturbing documented evidence — about a real criminal who is dead, a co-conspirator who is serving a federal sentence, and a justice system that still has not delivered accountability to the survivors he harmed. That real story does not need fabrication. ^[26]