Fake Trump Audio About Epstein Files Circulates Online After AI Tool Release

In a striking demonstration of how artificial intelligence can be weaponized to create convincing misinformation, a fabricated audio recording purportedly featuring President Donald Trump has been making rounds on social media platforms since early November.

The counterfeit recording depicts Trump supposedly panicking about the release of files related to convicted sex offender Jeffrey Epstein. In the fabricated clip, a voice mimicking Trump can be heard saying, “[We’re] not releasing the Epstein files! F Marjorie Taylor Green. I don’t care what you do. Start a fing war. Just don’t let ’em get out. If I go down, I will bring all of you down with me.”

The recording spread rapidly across Facebook, Instagram, and TikTok, with many users sharing videos that included visuals supposedly showing White House interiors and staffers. Some posts garnered significant engagement before being identified as fraudulent.

Digital forensics experts have confirmed the audio is entirely artificial, created using OpenAI’s recently released Sora 2 technology. The AI tool, which debuted on September 30, represents a significant advancement in synthetic media creation, featuring audio generation capabilities for the first time.

“This is exactly the type of misuse that technology companies and policymakers have been warning about,” said Dr. Claire Wardle, an expert in digital disinformation at the Harvard Kennedy School, who wasn’t directly involved in investigating this particular case. “The timing, just weeks after a major AI audio tool release, demonstrates how quickly bad actors can exploit new technology.”

Tracing the source of the fabrication, investigators found watermarks on a Facebook video featuring the fake Trump call that pointed to TikTok user @fresh_florida_air and Sora 2 user @bradbradt31. The @fresh_florida_air account has published numerous AI-generated videos, including the Trump clip in question, which was posted on November 16.

The same TikTok account previously shared a similar AI-generated video on November 5, featuring fabricated audio of Trump saying, “I don’t f***ing care how long this shutdown lasts. We will not lose to the Democrats. We will not release the Epstein files. I don’t care if the entire country starves.” That clip ends abruptly with Trump supposedly saying, “SNAP benefits can go to,” without completing the sentence.

The account has created multiple other politically charged AI videos, including one depicting four shirtless men wearing “MAGA” hats drinking alcohol while boiling in a pot over a campfire, discussing their desire to donate government checks back to Trump’s “defense fund.”

The rise of such convincing forgeries comes at a particularly sensitive time in American politics. Fact-checking organizations have been struggling to keep pace with the volume of AI-generated content appearing across social platforms.

“The democratization of AI tools means we’re entering an era where seeing—or hearing—is no longer believing,” said Paul Barrett, deputy director of the NYU Stern Center for Business and Human Rights. “This creates unprecedented challenges for voters trying to make informed decisions based on reliable information.”

Social media companies have implemented various measures to combat AI-generated misinformation, but identifying and removing such content remains challenging, especially when it spreads across multiple platforms simultaneously.

While this particular fake was relatively easy to identify due to its outlandish nature and the presence of watermarks, experts warn that more subtle AI forgeries could prove significantly more damaging to public discourse.

The incident highlights ongoing concerns about election integrity and information reliability as AI generation tools become increasingly sophisticated and accessible to the general public.