Rassemblement National uses
Sensationalizing Generative AI Imagery
in its EU Electoral Campaign.

Authors:

Salvatore Romano, Miazia Schueler, Denis Teyssou, Natacha Farina Groux, Sonia Grillot.

Affiliations:

Introduction

On 6th of June 2024, shortly before the 2024 EU parliamentary election, academics, journalists and civil society organizations such as VeraAI, Demagog and AI Forensics came together at the Media Studies Department at the University of Amsterdam for a so-called “Verificathon” workshop. This workshop aimed to answer a critical question that arises in contemporary electoral contexts: can the rise of generative artificial intelligence (AI) imagery change political campaigning as we know it?

The verification of AI generated content, specifically in an electoral context, is crucial to ensure information accuracy and transparency for fair democratic outcomes. Through our verifications, we intend to uncover the online dissemination of political misinformation, which can significantly impact voter perception of political landscapes. We believe that it is not only the voter’s right to be served reliable and authentic political information, but also the voter’s right as a user of various social media platforms to be made aware of the type of political content they are consuming online. Thus, investigations into AI generated content attempt to ensure that the dissemination of political information during electoral periods uphold ethical communication standards for the sake of electoral integrity.

This report focuses on the Rassemblement National’s political campaign, the French right wing party. We uncover their extensive usage of undisclosed Generative AI in their online campaigns and revolve around sensationalistic imagery to spread propaganda during the electoral period. Additionally, none of this AI-generated content has been flagged as such by social media platforms that are co-hosts of their electoral campaign, and the content has not been watermarked by who generated it.

Methodology

To detect the dissemination of AI-generated content online, we started by looking at material shared by official websites and social media accounts of various politicians and parties running for the EU parliamentary Elections scheduled between 6th and the 8th of June. We decided to focus on the French right wing party Rassemblement National as a case study, since the usage of Generative AI seems to be used consistently across european far-right parties, as seen with the AFD in Germany and with Lega Salvini Premier in Italy.

We started by analyzing the official website “rassemblementnational.fr” and collected all the social media channels it listed. As seen in Table 1, we found that only the official party channels and the ones of two main candidates are linked.

Table 1: The channels linked in the official website of the Rassemblement National.

We then started to look at the content shared on social media by those channels from the last 3 months, to limit the scope of this research to the electoral campaigning period. We scoped the content for AI generated images or videos.

To fact-check the images found we relied on VeraAI, a collection of software tools for Generative AI detection and fact-checking of information integrated into the InVID -WeVerify verification plugin.

To expand the list of official channels related to the party, we did a third level snowball sampling, collecting all the links contained on the Youtube channels present in Table 1. We first analyzed the Youtube videos’ descriptions, and the Instagram posts’ descriptions, without finding any additional link. We then moved to the official Facebook pages, and added to the list of URLS the one containing all the videos posted by the Party account Rassemblement National, and we focused on the ones posted in the last 3 months.

In table 2, you can see all the links related to the the “L’Europe Sans Eux” campaign:

Table 2: the list of the channels related to the “L’Europe Sans Eux” campaign
and linked in the official campaign website.

Together with these links, the website also provides links to directly create a post and re-share the official Bardella’s webpage vivementle9juin.fr on Linkedin, Telegram and WhatsApp.

Finally, we collected all the images present on the webpage https://leuropesanseux.fr, and we analyzed them. We then checked how these GenAI images were shared across the socials and whether they have been flagged by the content creators or the platforms as AI generated images.

Findings

Specticalized AI-Generated Imagery Promoting Far-Right Narratives

The Rassemblement National’s initiative L’Europe Sans Eux promotes political imagery that is mainly AI-generated. This is an integral part of their campaign, shared across all visual social media channels such as Facebook, Instagram, TikTok and X.

We started analyzing the most popular content on their Facebook official page according to CrowdTangle, and we used the InVID -WeVerify keyframes component to fragment the video and search by reverse similarity to find where those images appear. We found a visual montage depicting Macron with Ursula von der Leyen with a red cross and the mention of the motto “L’Europe Sans Eux” (Figure 1)

Figure 1: The main logo of the campaign “L’Europe Sans Eux”,
found at the end of an official FB video.

By similarity, the above image was retrieved in several communications of Rassemblement National on Facebook but also on Twitter, including in videos. Taking a snapshot of Mr Macron’s face, we passed it to the VeraAI synthetic image detector and found it was detected as AI-generated.

Figure 2: The veraAI detector of the InVID -WeVerify plugin finds strong evidence

suggesting that this image is synthetic

Interestingly, this GenAI content is also used in some videos spread on the official accounts, mainly when they make more sarcastic comments about political opponents or mocking the power.

Examining further the website https://leuropesanseux.fr, we found that other images contained in the page were Generated with AI. We then decided to take a deeper look at this campaign website and all its social media accounts, and make it the focus of our report.

More specifically, the imagery is designed to denigrate the duo Emmanuel Macron and Ursula von der Leyen depicting them as the main actors responsible for the immigration and energy crises, as well as the presumed islamisation of Europe. Those dramatized memes use text in big blocks and background images generated by AI.

Figure 3: The veraAI detector of the InVID -WeVerify plugin finds very strong evidence suggesting that this image (retrieved from Twitter) is synthetic.

The GenAI nature of that image can also be detected visually mainly on the people's faces in the background. The same image is also shared on Facebook and has been the second most shared post on a data analysis made on “leuropesanseux” on CrowdTangle.

A variant on the same topic is posted on the website and reuse on the official party account on Facebook :

Figure 4: The veraAI detector of the InVID -WeVerify plugin finds strong evidence

suggesting that this image is synthetic

Again the GenAI nature of that image can also be detected visually mainly on the people's faces in the background.

The following image was posted by @leuropesanseux on their X account on May 2:

Figure 5 : the VeraAI is not conclusive on this image published on Twitter in May 2024 but the visual exam shows blurred faces characteristic of GenAI images

Figure 6: A zoom of Figure 5.

On the Facebook, Twitter, Instagram or TikTok videos where those images have been used, we never found any mention of being AI-generated. The images themselves do not carry any logo or explicit mention of the political party.

Moreover, the comment sections on social media platforms such as Instagram and X, formerly Twitter, show little to no proof of critically questioning the artificially-generated visual content. Instead, comments lean into the far-right messaging by affirmative and extremist commentary. The debate seems to be very polarized into two different opinion groups, but there is no discussion on the technology used to create the pictures.

Platform moderation of the Generative AI imagery

While this content is freely shared online, and has been proven to be AI-generated, we found that none of these posts are flagged as Generative AI content on social media channels such as Instagram, TikTok and X. The “AI” or “Made with AI” label is an automated feature Instagram has implemented for both automated detection and manual flagging of posts. Similar to TikTok, it is mandatory according to the platform’s rules to disclaim if a content is generated with AI or not, and none of the videos containing this type of content has been declared as such by the Party. Here, for example, the AI generated propaganda gets almost 2000 likes in one Instagram post. Here a similar post gets 1600 views on X 1, and here is one of the many posts on TikTok containing some AI generated images like the logo of the campaign.

This shows a critical lack of platform content moderation and accountability for the sensitive and far-right content they host. It fosters a potentially distorted electoral process, where users and voters are influenced by unchecked and potentially harmful AI-generated propaganda.

This underlines crucial shortcomings of social media platforms in their content moderation strategies to identify AI-generated content. The inability to rigorously flag such content allows the spread of potential misinformation and undermines the efforts to maintain a transparent and authentic political communication in the digital sphere. This is especially crucial for political campaigning in electoral periods and compromises the transparency and authenticity of information shared and poses a strong risk for the integrity of the 2024 EU parliamentary elections.

Discussion

The Rassemblement National’s initiative L’Europe Sans Eux promotes political imagery that is almost exclusively AI-generated. This is an integral part of their widespread campaign, shared across all visual social media channels such as Facebook, Instagram, X. Interestingly, the AI generated content was predominantly shared on the "L'Europe Sans Eux" campaign accounts, and rarely shared on the official accounts of Rassemblement National, Marine Le Pen and Jordan Bardella accounts (see example here). Most prominent party candidates link to the campaign on their social media channels, while not explicitly sharing the campaign’s content. This shows a distinct top-down politically-strategic initiative carried out by the party itself.

More specifically, the imagery is curated to show dramatized right-wing narratives through aesthetic choices as a means of exaggerated storytelling tactics, and magnifies issues through a limited lens. Hence, this spectacularization of narratives depict factually misleading imagery, intended to enforce radicalized ideologies.

This study highlights how modern political campaigns, especially those with far-right ideologies such as the Rassemblement National, leverage advanced technologies such as generative AI to amplify their radical narratives and influence public opinion. The lack of critical engagement from the audience is indicative of potential vulnerabilities in the public’s ability to discern fact from fiction, which can lead to the normalization and spread of extremist ideologies. This underscores the importance of verifying AI-generated content and promoting media literacy to ensure a well-informed electorate and uphold the integrity of democratic processes.

The lack of effective content moderation and labeling of AI-generated posts on social media platforms like Instagram risks spreading misinformation and undermines the integrity of the 2024 EU parliamentary elections by allowing unchecked and potentially harmful propaganda.

Moreover, posting non-watermarked AI images violates commitment 3.b of the 2024 European Parliament Elections Code of Conduct, which is a voluntary, non-binding agreement signed by all European Groups, but not directly by the Rassemblement National. In general, spreading AI generated images without clear disclosure of their origin for sensationalistic reasons, might be very dangerous in the context of elections, since it could mislead voters. The fact that no comment in the posted content points out the Generative AI origin of the content, let us think that this is the case.

Conclusions

• Some politicians are using GenAI content at scale for the first time during EU elections

• RN created an ad-hoc website for that, that is not the main one, but it is still linked to all their campaign material. RN official accounts are tagged in the GenAI content shared on social media platforms (particularly on Instagram and Facebook), showcasing a strong connection.

• Platforms did not effectively (if at all) detect and flag/remove this content.

• The GenAI imagery is designed mainly to denigrate the duo Emmanuel Macron and Ursula von der Leyen depicting them as the main actors responsible for the immigration and energy crises, as well as the presumed islamisation of Europe.
  
  
  

Appendix

Here we add a non-exhaustive list of the AI generated images we found related to Rassemblement National.