Beware of fake images created by AI!
This spring, a certain Amandine Le Pen with long blonde hair was all the rage on the social network TikTok. This young woman in low-cut clothing presents herself as the niece of the leading figure of the National Rally (RN). The resemblance is striking, the exposure for the party maximum. Like this publication where she poses in an evening outfit, with a comment: “The dress I'm going to wear for Jordan's victory. » In reference to Jordan Bardella, the RN candidate leading the polls for the European elections on June 9.
Only problem, the Le Pen family does not include any Amandine. In reality, these larger-than-life videos, viewed millions of times before being deleted, were created from scratch. This is a “ deepfake » – video and audio recording made, modified and made credible using artificial intelligence (AI). This is the first time that such manipulation of reality has been involved in a campaign. THE deepfakes abound on the networks, at the risk of disturbing the 360 million European voters.
Joke or interference?
Internet users were thus able to discover Jordan Bardella in the middle of a video game where the goal would be to make migrants fall into the water… In reality, if the far-right candidate was indeed filmed with a controller in his hands, However, he was playing a game of… football. The head of the Renaissance list, Valérie Hayer, was also targeted, with a very cheesy fake clip generated by AI which created a buzz on the Web. “How could anyone vote for people with such a ridiculous presentation? » denigrates an anonymous person. The discredit has been cast. “They want to keep the little music of a worthless and unknown candidate alive. This can work on people who briefly watch the videos,” responds a Renaissance campaign manager.
From ill-intentioned jokes to foreign interference, it is difficult to trace these content creators. “I have the impression that new deepfakes every day, especially on the personalities of the majority,” he continues. Recently, a false announcement attributed to the Prime Minister caused a scandal among young people: he would have banned damaged haircuts at school. Gabriel Attal had to deny it.
“Previously, all family members watched the television news together. However, in today's society, information is consumed individually, there is no longer a filter that allows one to put this or that statement into perspective,” explains Nicolas Arpagian, vice-president of the cybersecurity consultancy Headmind Partners. Without hindsight, some fall into the trap.
“This content is used for destabilizing purposes to mix serious subjects with bickering or worse, to create disorder on the political scene,” observes Guillaume Kuster, co-founder of CheckFirst, which fights against disinformation on social networks. Slovakia will not say the opposite. In September, 48 hours before the legislative elections, a deepfake audio of progressive leader Michal Simecka where he claims to want to rig the election sows discord. Subject to the right of reservation provided for by law, the candidate was unable to refute the content and saw the victory slip away to the benefit of the populist pro-Putin party. With often little-known candidates, European elections lend themselves to manipulation.
Deception for voters
In order to avoid a Slovak-style scenario, the European Commission has equipped itself with a legislative arsenal. Entering into force last February, the Digital Services Act (DSA) urges digital giants to hunt down deepfakes which can “mislead voters”, explains a Brussels official. But while waiting for these safeguards to be put in place, Internet users are invited to exercise discernment. “In the past, proof through images was conclusive. Today, with AI which plays on cognitive bias, we only see what confirms our convictions,” analyzes Nicolas Arpagian.
The situation becomes complicated when the candidates themselves resort to rigging, like Juliette de Causans (Ecology list in the center), who manipulated her campaign poster: the forty-year-old appears in other guises and looks like a model . “As Arcom only gives us thirty seconds of speaking time, all that remains is the photos: it’s called democracy,” she brazenly justifies on X (formerly Twitter). Did you say intelligence?
