French Elections: Far Right Uses AI to Support Its Political Messages

This article was originally published in English

European far-right parties have used images generated by artificial intelligence (AI) to support their political messages, according to two new reports, but it is difficult to assess the impact these images may have had on recent French and European elections.

The non-profit organization AI Forensics crawled the social media pages of 31 French political parties to determine whether they could spot AI-generated content (genAI) between May 1 and June 28, during the European elections and the first round of the French legislative campaign.

They then submitted these images to AI-related search engines to verify that an AI tool had created them.

In total, the team discovered 51 cases of generative AI images used by French parties, with the National Rally (RN), Reconquest and Les Patriotes being the most involved, but the researchers caution that this is a conservative estimate.

“This is the first electoral cycle where (generative) AI exists, but we are sure that in the coming years it will be used more and more”said Salvatore Romano, a researcher at AI Forensics, Euronews Next.

“If we can’t manage this now, when the scale is so small, it will be harder to manage when the technology is indistinguishable from the actual content,” he added.

The far-right RN party came out on top in the first round of the French elections last Sunday, with 33% of the vote, and in the European elections in June, where it obtained 31.5% of the vote.

Where has AI-generated content emerged?

Most AI-generated campaign images used by French parties “convey messages of anti-immigration, insecurity in France, lack of freedom of expression, a defective judicial system and weak protection of agriculture”according to the AI ​​Forensics report.

“The images are being used as a fear-mongering technique to say that France is not safe in the EU and that it is preying on vulnerable people in our society.”added Miazia Schüler, researcher at AI Forensics.

These images are often “hyperrealistic”meaning the average user can clearly see that they were created by AI, Schüler explained.

Others were more subtle, like imitations of political cartoons found in newspapers.

Some parties, such as the far-right Reconquête party, have also used generative AI to create images of French President Emmanuel Macron and European Commissioner Ursula Von Der Leyen.

During the European campaign, the RN created a website called “Europe without them” in the run-up to the election, which also called into question the leadership positions of Macron and Von Der Leyen, according to a study by the Atlantic Council’s Digital Forensic Research Lab (DFRLabs).

The website, which is still online, contains at least three AI-generated images: petitions against “EU’s contempt for people, mass immigration and promotion of the hijab”according to the report.

The tip of the iceberg

Valentin Chatelet, a researcher at the Atlantic Council, said he was “difficult” to know what impact this type of AI-generated content had on the RN’s success in the European and French elections.

“I think it has a positive impact on the material that the party uses to convey a particular idea, such as Mr. Bardella’s particular position on an issue.”Mr. Chatelet said.

“(AI-generated) images are also used to shock people… until they become mainstream”he added.

Euronews Next contacted the National Rally (RN), the Identity and Democracy group, the Patriots and Reconquest, but did not receive a response.

RN candidates have been receiving training in AI, including how to use ChatGPT since January 2024, according to French broadcaster BFMTV. The training booklet gives examples of how AI can help political candidates, such as creating campaign posters and researching arguments for debates, the broadcaster reported.

Neither the AI ​​Forensics study nor the Atlantic Council’s DFRLabs study looked at candidates’ personal messages or content created by party fanbases.

Salvatore Romano believes that their study is only “the tip of the iceberg” regarding the amount of unlabeled election content circulating on social media platforms.

Serious negligence of technology companies in the use of AI

The AI ​​Forensics report found that none of the AI-generated images were labeled as such.

“Our findings reveal serious negligence on the part of political parties and platforms in failing to meet their commitments to label AI-generated images in their political campaigns.thus highlighting the urgent need for transparency,” the report reads.

Major tech companies including Google, OpenAI, Meta and TikTok signed an open technology agreement in February that “sets expectations for how signatories will handle misleading election content generated by AI” on their platforms before the European elections.

The first commitment of this agreement is that the platforms “(identify) realistic images generated by AI and/or (certify) the authenticity of the content and its origin”, can we read in a copy of the agreement .

Tech giants have also promised to review their models in order to “understand the risks they may present” for misleading content generated by AI.

Some political parties have signed their own version of this agreement with the European Commission, which has committed to “preventing deliberate deception of the public, including through the use of artificial intelligence”we can read in a copy of the agreement.

The Identity and Democracy (ID) group of the European Parliament signed this code of conduct, of which the RN is a member. The other French national parties did not do so.

Salvatore Romano said AI Forensics would send its report to the European Commission in the hope that there would be some form of repercussion against parties that violate these codes of conduct.