As U.S. election race heats up, AI-generated images proliferate on social media

On September 1, 2024, U.S. Vice President Kamala Harris posted a graphic (archived here) on X (formerly Twitter) that read, "Donald Trump vows to be a dictator on day one." In response, Tesla CEO and X owner Elon Musk, who has publicly endorsed former President Trump, shared an AI-generated image of Harris wearing a red jacket and hat featuring the communist hammer and sickle logo. The image was not labeled as AI-generated, leaving its satirical intent unclear. Since Musk's post, several users have shared the image, many of whom seemed to believe it was real.

Screenshot of the X post by Elon Musk. (Source: @elonmusk/X/Modified by Logically Facts)

Multiple AI-generated images related to the upcoming U.S. presidential election have been shared on X, ranging from satire to disinformation. For example, Trump, reinstated on X after Musk's takeover of the platform in October 2022, recently posted an AI-generated image of Harris addressing a crowd with communist flags in the background.

Image shared by Donald Trump on X. (Source:@realDonaldTrump/X/Modified by Logically Facts)

Other such viral images include doctored photos of Trump kissing Musk while dressed in Nazi uniforms and Harris in communist attire. These images feed into narratives questioning the political ideologies of both Trump and Harris. Trump has faced accusations of using Nazi rhetoric, while Harris has been falsely labeled a communist by Musk and others on social media.

Images of Donald Trump and Elon Musk shared on X. (Source:@YourAnanNews/X/Modified by Logically Facts)

Platform policies and inconsistent enforcement

Platforms like Meta have policies requiring AI-generated images to be labeled, often automatically flagging them if detected. 

X also has a policy against misleading synthetic media, stating that users "may not share synthetic, manipulated, or out-of-context media that may deceive or confuse people and lead to harm." However, X allows satire and memes if they don't cause "significant confusion about the authenticity of the media." The enforcement of this policy under Musk’s leadership has been inconsistent.

These concerns come to the forefront as the 2024 presidential race heats up, with Trump and Harris facing off to become the 47th president of the United States.

Targeted disinformation using AI

AI-generated images are increasingly being deployed to target specific voter groups. For example, Trump supporters have created AI-generated content aimed at Black voters, attempting to sway their support towards Trump. The BBC has identified numerous such images depicting Black voters endorsing Trump, highlighting this trend as a significant issue ahead of the November election.

On Truth Social, Trump posted AI-generated images (archived here) featuring Taylor Swift fans wearing "Swifties for Trump" apparel, including one image that falsely showed Swift herself endorsing Trump. Swift has not endorsed Trump in 2024 and had supported President Joe Biden over Trump in 2020. Trump later admitted that these images were fabricated, attributing their creation to others.

The Truth Social post by Donald Trump. (Source:@readDonaldTrump/Truth Social/Modified by Logically Facts)

This form of disinformation obscures the truth during a crucial election period. Trump has a history of disseminating falsehoods and conspiracy theories, including unfounded claims about the 2020 election. Recently, he falsely accused the Harris campaign of using an AI-generated image of a rally, which was actually a genuine depiction of a large crowd.

Concerns about AI's role in voter manipulation are heightened this election cycle. Earlier this year, an AI-generated fake robocall impersonating President Biden and urging Democrats not to vote in the New Hampshire primary had sparked significant alarm.

What about safeguards?

As disinformation continues to spread through AI-generated content, some platforms are taking steps to prevent the misuse of their technologies. Microsoft’s Bing image generator, for instance, blocks prompts involving political figures like Trump, Harris, and Musk, citing policy violations.

Microsoft Copilot Image Generator refusing to generate an image with the prompt ‘Kamala Harris’. (Source: Logically Facts)

Midjourney and OpenAI's DALL·E both restrict the creation of images for political purposes. Midjourney's policy prohibits generating images for political campaigns or election influence, while DALL·E bans content related to politicians, ballot boxes, and protests. Additionally, both ChatGPT and Midjourney have reportedly banned specific names, including ‘Kamala Harris’ and ‘Donald Trump’.

However, despite these measures, a March 2024 report by the Center for Countering Digital Hate found that platforms like Bing and Midjourney failed to restrict the creation of images containing disinformation in 41 percent of cases.

Following a controversy over offensive AI-generated images in February, Google recently announced it would resume generating images of people and emphasized that it is “applying election-related restrictions” to its generative AI products, including Search AI Overviews, YouTube summaries, and image generation in Gemini. 

In contrast, Logically Facts found that Grok, an AI chatbot developed by xAI and accessible to X Premium users, still generated an image with the prompt “Donald Trump in a communist uniform with a hammer and sickle,” highlighting a gap in enforcement.

Screenshot of an image of Donald Trump generated by Grok. (Source: Grok/Modified by Logically Facts)

An August 2024 NPR report revealed that Grok could generate images depicting ballot drop boxes being stuffed and Harris and Trump holding firearms. Additionally, several social media users have reported creating AI images of Trump, Harris, and Obama using Grok.

The Center for Countering Digital Hate has raised concerns about Grok’s lax safeguards and its potential to produce misleading election-related images. Despite these issues, Grok lacks consistent guardrails and produces varying results with each attempt.

How to detect AI-generated images

As concerns about misleading AI-generated content grow, it's essential to know how to identify these images. Even though such images have become quite realistic, there are still some clues that could help you distinguish between a real image and an AI-generated one:

1. AI often struggles with details like human fingers and hair edges.

2. Look for unnaturally smooth skin textures, repeated patterns, or overly symmetrical features.

3. AI-generated images sometimes distort text and logos.

4. Some AI image generators include metadata in the images that can help identify their origin.

5. Check comments on social media posts for clues about the image’s authenticity.

6. Lastly, always question the source of the image and the information shared with it.

Remember, no method is foolproof. A well-made AI image can still be very convincing, so it's essential to verify before sharing.