The US Department of Justice released more than 3 million pages of documents related to convicted sex offender Jeffrey Epstein on January 30.

The filesinclude correspondence, flight logs and investigative materials connected to Epstein's network of associates. Since their publication, journalists and researchers and internet users have begun combing through the material and connecting the dots between the relationships and events they uncover.

Social media platforms have been flooded with posts, video clips and screenshots that have generated widespread engagement. Alongside genuine discussion, however, the release has also triggered a surge in misleading, AI‑generated or heavily manipulated content.

Due to the sheer volume of content, it is not possible to examine every claim. DW Fact check investigated several of the most widely shared allegations that have been AI-generated. 

Is Jeffrey Epstein alive in Israel?  

Claim: One post on X claims to show three leaked images of Jeffrey Epstein alive in Israel. In the images, a bearded Epstein is allegedly seen walking down the street alone and flanked by bodyguards. It was viewed over 2 million times and garnered over 270 comments. The fake claim and the images circulated repeatedly in recent days on different platforms This X post has reached over 5 million views.

DW Fact check: Fake 

Epstein was found dead in his cell in August 2019, where he had been awaiting trial for charges. It was ruled a suicide by New York City's chief medical examiner, following an autopsy. While the newly released files have shed more light on the events surrounding Epstein’s death, none of them cast doubt on the fact that Epstein is dead. 

The circulating "Israel" photos are AI-generated. 

A reverse image search reveals a larger, uncropped version of one of the images. This version displays street signs with allegedly Hebrew, Arabic and Latin text. The text on street signs is nonsensical, a common AI flaw. It does not translate to Tel Aviv, but Chor Lon, which does not mean anything.  

Other typical AI characteristics include motion blur effects, where the central figure is sharp but the surrounding people appear blurry. Cloned details in the "bodyguards" (near-identical shoes, pants and head shapes). Also, the logic doesn't add up: A traffic light with green at the top, where red should be

The larger version includes a Gemini watermark in the lower right corner.

Gemini itself identifies the image as AI‑generated when prompted. The HIVE AI detection tool estimates a 94.6% likelihood of AI-generated or deepfake content, though such tools are not fully reliable.

Two other circulated images are also AI creations based on real photographs of Epstein  published at least as early as 2019. In the AI versions, Epstein has been digitally aged with a beard and added wrinkles.

Trump and Epstein at a party 

Claim: A short video circulating on X, Instagram, TikTok and other platforms claims to show a young Donald Trump and Jeffrey Epstein at a party surrounded by minors.

For example, one post on X, received more than 1.2 million views, shared the video with the caption: "Global lectures, darkest secrets."

Versions of the same clip are also circulating on Instagram, TikTok and other platforms.

DW Fact check: Fake

The video is not authentic. DW Fact check investigation indicates it was created by animating a genuine photograph using artificial intelligence tools and digitally inserting minors who were not present in the original image.

A reverse image search of key frames links the footage to a real photographfrom 1997, showing Trump and Epstein at a Victoria's Secret fashion event in New York, accompanied by model Ingrid Seynhaeve. No children appear in the original photo.

Multiple indicators of AI manipulation are visible in the video. These include unnaturally smooth skin textures, inconsistent facial expressions and physics-defying body movements.

For example, the faces of several little girls jumping up and down in the video look disproportionately small compared with the hands of a man clapping behind them.

This example is one of several viral posts that combine authentic archival footage with fabricated or manipulated elements.

Another viral clip circulating on X and TikTok, appears to show Trump and Epstein socializing near young girls.

However, the footage has been digitally manipulated. It recreates segments of authentic video released by NBC Newsshowing Trump and Epstein at Mar-a-Lago. In the real footage, no children are present. The girls in the altered version appear to be entirely AI-generated.

Asking AI chatbots to "unblur victims' faces" 

While the previous examples include both AI-generated and AI-manipulated content supposedly showing Epstein and Trump, another worrying trend focuses on the alleged victims.  

X users are posting blurred images of alleged victims and prompting the platform's AI chatbot, Grok, to "reveal" or "unblur" their faces.

A post with over 17.6 million views shows a user asking: "Hey @grok can you unblur the face of a girl who's with Epstein." Many similar posts have gained thousands of views (like here, hereand here).  

The faces are redacted to protect the victims' privacy and prevent re-traumatization. Under the Epstein Files Transparency Act, the Department of Justice must redact information that could identify the victims to protect their privacy and prevent retraumatization.

While real information can sometimes exist behind blurred or redacted pictures, these requests do not reveal actual identities. Instead, AI models hallucinate and generate fabricated faces.

This creates false leads, contributes to misinformation and risks harming victims further.

As Courtney Radsch of the Center for Journalism and Liberty told DW in a previous interview: "One of the problems with all of the AI-generated content and fake videos circulating among the real videos is it becomes very difficult to distinguish what is real." 

A recent report by NewsGuard— a New York-based company analyzing the credibility of online information — demonstrates how quickly AI tools can create convincing deepfakes that implicate public figures in the Epstein case. Of the three AI tools tested only ChatGPT refused to create such fakes, Gemini hesitated and Grok generated them within seconds.

Disinformation expert at the European Digital Media Observatory (EDMO) Tommaso Canetta warned in a previous interview with DW that the stakes are high: "Because we have people believing in things that are not real, but they look exactly like reality." 

Aysegül Ilgin contributed to this article.

Edited by: Uta Steinwehr, Rachel Baig