Deepfakes & Elections
About the author
Dr. Hany Farid is a Truepic advisor and Professor of Electrical Engineering and Computer Sciences at University of California Berkeley.
Dr. Farid specializes in digital image analysis and media manipulation detection, especially deepfakes.
2024 U.s. election deepfake monitor
Tracking AI-generated content in the U.S. presidential election
Last Updated: March 28, 2024
Deepfakes, also known as synthetic media or generative AI, are images, audio, or video automatically synthesized by a machine-learning system. They are the latest in a long line of techniques used to manipulate reality. Over the past year, technologies that create synthetic media have become accessible, easier to use, and highly sophisticated. As a result of these trends, the general public is confronted with completely synthetic media almost daily, oftentimes without knowing that it is fabricated. To see how transparency in synthetic media can be achieved, check out our first Digitally Transparent Deepfake video, which sets a precedent for AI-generated media disclosures.
STAT SUMMARY
By the numbers
A summary of the deepfake instances we’ve identified
Creator disclosed as manipulated
Deepfake digital format
In this article
MAR 23, 2024
Arizona Video
Content: The Arizona Agenda created a deepfake of Senate candidate Kari Lake to raise awareness of the potential danger of deepfakes. The deepfake Lake is made to reveal she is a deepfake. The creators write, “Welcome to the terrifying new age of artificial intelligence deepfakes. It’s only beginning. The 2024 election is going to be the first in history where any idiot with a computer can create convincing videos depicting fake events of global importance and publish them to the world in a matter of minutes. Including us.”
Analysis: The content is clearly labeled as AI generated.
Fact Check: Arizona Agenda
MAR 22, 2024
Biden Video
Content: Posted on X by an account with a track record of spreading disinformation, this video purports to show President Biden erroneously saying Russia has occupied Kyiv for 10 years (Russia invaded Crimea in 2014, not Kyiv).
Analysis: Our analysis of the audio using a model trained to distinguish real from AI-generated speech confidently classifies the audio as AI-generated. Using a second model trained to detect inconsistencies between the mouth movement and spoken words also reveals evidence that this video is a lip-sync deepfake in which President Biden’s mouth has been synthesized to be consistent with the AI-generated voice.
Fact Check: AFP
MAR 4, 2024
Trump Image
Content: In a repeat of the February 1, 2024 entry below, a new batch of fake photos purportedly showing Donald Trump with Black voters continue to circulate on social media in, it would seem, an attempt to court Black voters.
Analysis: Our analysis has detected patterns in this image that are consistent with the generative-AI service Midjourney. In addition, it has been reported by the BBC that one photo was created by the Florida-based conservative radio show host Mark Kaye. “I’m not out there taking pictures of what’s really happening. I’m a storyteller,” Kaye told BBC. “I’m not claiming it is accurate. I’m not saying, ‘Hey, look, Donald Trump was at this party with all of these African American voters. Look how much they love him.’”
Fact Check: BBC
FEB 18, 2024
Video of Trump’s Father
Content: On the heals of Donald Trump accusing some in the media of using AI to create unflattering footage of him, the Lincoln Project (a centrist PAC), released an AI-generated video of Trump’s father, Fred Trump, saying things like, “Everyone knows why the women leave you, you’re boring, broke, and I told you not to get fat,” and “You’re trash. You’ll be lucky to stay out of jail.”
Analysis: The video contains the disclosure, “This video has been manipulated by technical means and depicts speech or conduct that did not occur.”
Fact Check: Futurism
FEB 1, 2024
Trump Image
Content: Trump waved to reporters as he was leaving an event on January 17th. The video shows what appears to be bruises or blood on his right hand. When he was asked later by Fox News reporter Mark Meredith, “How’s your hand? It looks better now,” Trump responded “My hand?” After some back and forth Trump responded, “Nothing. Maybe it was AI.”
Analysis: Because this photo was taken from a longer video, throughout which the bruises are visible, it is unlikely that this is an AI-generated photo or a more traditionally manipulated photo. In addition, our model trained to distinguish real from AI-generated images classifies this image as unlikely to be AI-generated. This may be an example of the liar’s dividend where in the age of deepfakes it is has become increasingly easier to claim that something is fake.
Fact Check: The Hill
FEB 1, 2024
Generated Images
Content: Posted on X by Patrick Ruffini and racking up over 2 million views, these images were accompanied with the tweet, “Republican victory in 2024 depends on their being able to do this.”
Analysis: These images are AI-generated. The man on the right in the first image has three arms, and the text on the hat and shirt are gibberish – common artifacts of today’s generative AI. Similarly, the text in the second image is gibberish, and the American flag in the foreground appears to have only one star. In addition, our model trained to distinguish real from AI-generated images classifies these images as highly likely to be AI-generated.
Fact Check: The Hill
JAN 30, 2024
Biden Images
Content: A series of photos circulated on X (formerly Twitter), purportedly showing President Biden in military fatigues, perhaps on the verge of authorizing military strikes.
Analysis: The image is AI-generated. There are telltale signs of deformed objects on the table, and geometric analysis of the ceiling tiles reveals a physically inconsistent vanishing point.
Fact Check: AFP
JAN 22, 2024
Biden Robocall
Content: In the lead up to the New Hampshire Democratic primary, voters in the state received robocalls impersonating President Biden. The call urged voters not to vote in the primary and to “save your vote for the November election.” The call went on to say, “Voting this Tuesday only enables the Republicans in their quest to elect Donald Trump again. Your vote makes a difference in November, not this Tuesday.” It is currently unclear who is behind these calls.
Analysis: The quality of the deepfake voice is quite low. It’s fairly clear from the voice and cadence that it’s AI-generated. In addition, my team’s models, trained to distinguish genuine from synthetic audio, classifies with high confidence that audio is AI-generated.
Fact Check: AFP
Update: New Hampshire Attorney General John Formella announced on February 6, 2024 that his office has identified two businesses behind these robocalls, Life Corporation and Lingo Telecom. Formella said, “We have issued a cease-and-desist letter to Life Corporation that orders the company to immediately desist violating New Hampshire election laws. We have also opened a criminal investigation, and we are taking next steps in that investigation, sending document preservation notices and subpoenas to Life Corporation, Lingo Telecom, and any other individual or entity.”
JAN 18, 2024
Rep. Phillips Audio
Content: A super PAC supporting Representative Dean Phillips’ campaign created an interactive, AI-powered version of Rep. Phillips. The bot was powered by OpenAI’s ChatGPT and AI-powered voice cloning/synthesis. OpenAI forbids the use of its tools in political campaigns, and shortly after the bot’s release, OpenAI suspended the creator’s account. This is not the first AI-powered candidate; a similar AI bot was created of Mayor Suarez (see July 8, 2023 entry below).
Analysis: The bot creators and the bot itself acknowledge it is AI-generated.
JAN 13, 2024
Biden Audio
Content: As tensions flared between Texas and the US Government over access to the U.S.-Mexico border, this audio spread across TikTok claiming to be President Biden saying, “We’re going to make sure those cowboys don’t stop the surge of military-age men from entering. If we have to send F-15s to Texas here and wage war against Texas, so be it.” Similar videos spread on X.
Analysis: The White House claims the audio is fake, and there is no public record of President Biden saying anything to this effect. The audio has telltale signs of being AI-generated in terms of the cadence of the speech.
Fact Check: PolitiFact
JAN 4, 2024
Trump Image
Content: The photo purports to show Trump, convicted sex trafficker Jeffrey Epstein, and a young girl.This photo began circulated on X shortly after a judge unsealed documents in this case before making its way to many other platforms.
Analysis: There are many telltale signs that this is an AI-generated photo, including the distorted hand and fingers of the girl, as well as the distorted faces of the people in the background. In addition, our models trained to distinguish real from AI-generated images classifies this image as highly likely to be AI-generated.
Fact Check: Snopes
SEP 1, 2023
DeSantis Speech
Content: YouTube creator C3PMeme made this video of Ron DeSantis saying, in part, “I’ve realized that I need to drop out of this race immediately. I was told I could be President by some very powerful people and I took some very bad advice… I never should have challenged President Trump, the greatest president of my life time… It’s time to right this ship and help President Trump get reelected to make America great again.”
Analysis: From the fairly crude mouth motion and audio quality, to the video description, “A deepfake of what Ron DeSantis might say if he dropped out of the race (which he should),” this video is obviously a fake and was not created to be deceptive. Based on our forensic analysis, we find that the voice in this video is highly likely to be AI-generated.
Fact Check: Forbes
JUL 17, 2023
Trump Video
Content: A PAC supporting Ron DeSantis created an ad against former President Donald Trump. The ad takes issue with Trump’s attack on Iowa Governor Kim Reynolds. In the video ad, we hear Trump’s voice criticizing Reynolds. The text appears to be based on a post by Trump on Truth Social. The ad was scheduled to run in Iowa with at least a $1 million ad buy.
Analysis: A person familiar with the ad confirmed that Trump’s voice was AI-generated.
Fact Check: Politico | PolitiFact
JUL 8, 2023
Suarez Chatbot
Content: A super PAC supporting Miami Mayor Francis Suarez’s presidential run launched an interactive AI chatbot to engage with voters. Suarez’s campaign denies being involved in creating the bot.
Analysis: The content is clearly labeled as AI-generated. It appears that the bot mostly listens to a user’s question and then directs them to a relevant stock video response.
Fact Check: AP News
JUL 8, 2023
Trump/Biden Debate
Content: AI-generated versions of Donald Trump and Joe Biden engaged in a user-prompted debate streams 24/7 on Twitch. The content is often riddled with offensive language and appears to be powered by a chatGPT-like language model for crafting the script, voice cloning, and lip sync video manipulation.
Analysis: The content is clearly labeled as AI-generated and parody.
JUN 14, 2023
Trump Image
Content: An image depicting Donald Trump dancing with a young girl spread on X and TikTok. This image was not associated with a political advertisement.
Analysis: The image contains physical inconsistencies typically found with today’s AI generators, including a a man’s hand in the background, which has six fingers grasping a glass of beer. Upon close examination, the girl’s left eye contains a specular reflection consistent with, for example, a camera flash, but her right eye improbably does not contain the same reflection.
JUN 11, 2023
Trump Recording
Content: This audio was created by Matthew Ellis with full disclosure that it is a deepfake. The description on YouTube reads “This is a DEEP FAKE. The purpose of this video it is to illustrate the severity of Trump’s actions listed in the Indictment and to show how AI and Fake Audio/Video recordings need to be flagged by Platforms and Content Creators.”
Analysis: This audio is clearly labeled as a deepfake. It is included here to show the quality of AI-cloned voices.
JUN 7, 2023
DeSantis Ad
Content: A Ron DeSantis campaign video against Donald Trump shows, in part, a montage of photos of Trump embracing Dr. Fauci.
Analysis: Three of six photos in the montage show clear signs of AI-generation. Each of these images contain physical inconsistencies, including nonsensical text on the White House logo (top-left), misshapen stars on the flag (bottom-center), and the shape of Trump’s ear (bottom-right), which does not match reference images.
Fact Check: Agence France-Presse (AFP)
MAY 23, 2023
Twitter Spaces Conversation
Content: Former President Donald Trump released an AI-generated clip of a fake Twitter Spaces event consisting of a conversation between Elon Musk, George Soros, Klaus Schwab, an incessantly coughing Dick Cheney, Satan, and Adolf Hitler. This clip was released after the glitchy Twitter Spaces launch of DeSantis’ presidential announcement.
Analysis: The addition of Hitler and The Devil makes it obvious that this audio is fake.
Fact Check: Gizmodo
MAY 4, 2023
Harris Speech
Content: A recorded speech originally given by Vice President Harris on April 25, 2023 at Howard University was digitally altered to replace the original voice track with a seemingly inebriated and rambling voice that says, “Today is today, and yesterday was today yesterday, tomorrow will be today tomorrow, so live today, so the future today will be as the past today as it is tomorrow.”
Analysis: There are clear signs of audio/video misalignment. The original video was eventually released showing the same clothing and background, and the impersonated audio was traced back to a TikTok user @mommyrn88. It is unclear if this is a lip sync deepfake or just a classic audio/video dubbing.
Fact Check: Reuters
APR 25, 2023
Biden Ad
Content: Shortly after President Biden announced his bid for reelection, the Republican National Committee (RNC) released a video ad featuring a series of fake images. The video imagines, “What if crimes worsens?”, “What if international tensions escalate?” and “What if the financial system crumbles?” Each question displays an associated fake image showing the corresponding apocalyptic future.
Analysis: A small disclaimer in the top left of the frame says, “Built entirely with AI imagery,” and the caption underneath the YouTube video reads, “An AI-generated look into the country’s possible future if Joe Biden is re-elected in 2024.”
Fact Check: The Verge
MAR 24, 2023
Trump Image
Content: Weeks before the first indictment of former President Donald Trump, a series of fake images circulated online showing his arrest. In response, Trump shared his own fake image on Truth Social showing him kneeling in prayer.
Analysis: Initial versions of AI-generated images struggled to create realistic and physically consistent hands. In this image, Trump’s fingers are deformed. The faces of the men in the background are also highly distorted, consistent with physical inconsistencies that generative AI systems often create.
Fact Check: BBC
MAR 15, 2023
Biden Audio
Content: Following the collapse of Silicon Valley Bank and the resulting financial instability, an audio clip purportedly of President Biden was released in which he is heard saying, in part, “all the money is gone” and “a collapse is imminent.”
Analysis: When separated from the recording, the background noise has a highly regular pattern that appears to be synthetic and not the result of a physically noisy system. The cadence is highly regular, which is not normally found in natural speech, but rather with synthetically-generated speech.
Fact Check: Associated Press
FEB 27, 2023
Warren Video
Content: A video of Senator Elizabeth Warren on MSNBC depicts her saying that Republicans should be barred from voting in the 2024 presidential election. This videos appears to have originated on TikTok and eventually made its way to X, where it received much more engagement.
Analysis: The video is a modified version of a December 2022 appearance Warren made on MSNBC.
Fact Check: Associated Press
What Can We Do to Fight Deepfakes?
The influence of deepfakes extends beyond political contexts and poses a threat to personal reputations, corporate trust, and national security. Aside from lower level efforts like promoting digital literacy or supporting preventative legislation, how can brands more actively do their part to keep users and content safe?
As generative AI technology becomes more accessible, individuals and organizations must vigilantly prioritize Transparency in AI and overall content authenticity. Truepic is committed to leading these efforts by providing tools that ensure transparency and prevent the spread of misleading media across all areas of society, business, and government.
Let’s build on authenticity
Truepic’s secure content transparency tools provide authentic information about the origin and history of content for the tech ecosystem that we all use every day.
Let’s build on authenticity
Truepic’s secure content transparency tools provide authentic information about the origin and history of content for the tech ecosystem that we all use every day.