“A real threat”… Should we fear a campaign disrupted by deepfakes?

Emmanuel Macron as a yoga teacher. Éric Zemmour as a rapper in a cap. Marine Le Pen singing raï. Marion Maréchal disguised as Superwoman… It’s their voices, their faces, but it’s not about them. These fake videos generated using artificial intelligence, called “deepfakes”, are proliferating on social networks.

Most of the time, they aim to amuse Internet users. But this “hyperfaking” technology, which is making rapid progress, could well be used to try to bias elections by disseminating false information. A threat that hangs over the European elections, which will take place on June 9 in France, and which worries the political world.

Prominent examples in the United States

The American presidential election recently showed the (almost infinite) possibilities offered by these new technologies in the context of electoral campaigns. Last July, a supporter of Ron DeSantis broadcast in an advertising spot a sound clip of Donald Trump, then great rival in the Republican primaries. The voice of the former American president was actually cloned by an AI, the sentences in question having never been uttered. Two months later, the same Ron DeSantis appeared in a video to announce his withdrawal from the race for the White House. Problem: this was again a deepfake, massively relayed on social networks, which forced the candidate’s team to deny it.

This type of practice could spread in France, particularly ahead of the European elections. “It’s almost inevitable,” breathes socialist deputy Hervé Saulignac. “It is to be expected that some will use AI for disinformation or defamation. We can therefore imagine having deepfakes of excellent invoices which cause damage and bias the elections,” he adds. “There is a duty of extreme vigilance, because these tools are within reach of anyone. This is a real threat to our democracies,” also worries Constance Le Grip, LR MP.

Last February, an opponent of Joe Biden for the Democratic primaries admitted to having used AI to reproduce the voice of the American president in a telephone call. The message, addressed to thousands of Americans to convince them not to go to the polls, was produced in a few clicks, and for only $500…

Marine Le Pen’s fake nieces

France has so far been relatively spared from this type of controversy. The videos published on Instagram or TikTok mainly have fun making politicians sing or dance in a parodic manner. But recent examples have shown that the use of “entertaining” can also have a deeper impact. In mid-April, “influencers” presenting themselves as members of the Le Pen family burst onto TikTok, garnering hundreds of thousands of views. These sexy young women, borrowing the features of Marine Le Pen and Marion Maréchal, promoted the ideas of the RN and Reconquest. These false accounts, denounced by the entire political class, were finally deleted from the Chinese platform.

“The use of deepfakes raises questions about future campaigns. But we should not condemn everything either. There is something about entertainment, about humor. And the scenario of a false video published a few days before the election, which would give the impression of a declaration which was in reality never pronounced”, nuance Gaëtan Dussausaye, spokesperson for the RN, close to Jordan Bardella.

The flame party has still not digested the deepfake published by Loïc Signor at the beginning of January. The spokesperson for Renaissance had shared on his X account false wishes from Marine Le Pen speaking in Russian, a first within the French political class. The initiative had been denounced by the majority, proof of the concern surrounding the use of these new tools. “It wasn’t smart, we don’t need to use that to denounce relations between the RN and Russia,” regrets Erwan Balanant, MoDem deputy.

The risk of foreign interference

This controversy was all the more difficult as the government then sought to limit rigging practices through the Sren law (Draft law aimed at securing and regulating the digital space), transposition of the European text Digital Services Act (DSA), finally adopted on April 10. But the judicial machine sometimes seems slow compared to the speed at which false information spreads. “We have a certain number of measures put in place in recent years, but justice always intervenes after the fact. This is why we insist on the obligation of transparency of platforms, as part of self-regulation,” underlines Romain Rambaud, professor at Grenoble Alpes University and specialist in electoral law.

The European DSA text provides for fines of up to 6% of the platforms’ annual turnover in the event of failure to fulfill their duty of moderation. But the legislative elections in Slovakia at the end of September, which served as a test, have already shown the limits of the European system. The poll was disrupted a few hours before the vote by a false registration of a pro-Europe candidate. The extract, with unclear origins, was widely distributed on social networks before its deletion by these same platforms.

The action of the platforms in question

Under pressure from the European Commission, several tech giants, such as YouTube and Meta, have announced in recent weeks that they want to label any image generated by AI, in the run-up to several major elections in 2024. Meta therefore, the parent company of Facebook and Instagram, also met the teams of the European candidates in March, reports Samuel Lafont, director of digital strategy at Reconquête, to discuss this risk. “They have shown their credentials, the platforms know that they are playing very big on this point,” underlines this close friend of Eric Zemmour.

Because the war in Ukraine has accentuated fears of an attempt at destabilization from Russia. “What is most feared is the risk of a foreign threat to weaken this or that candidate with a deepfake, through the troll factory relay,” admits Anne Genetet, Renaissance MP. Socialist elected official Arthur Delaporte agrees: “The real danger is not the occasional use of a deepfake, but its use by foreign powers to destabilize the election.” The European elections in June will be a life-size test in this area, when 370 million voters are called to the polls.

source site