How Russia is using AI to influence the U.S. midterms ¦ You&AI

Elon Musk turns off Starlink, AI agents are naughty again, and the US midterms are coming, so Russia is of course doing some meddling. My name is Aisha Bajaj, and welcome to UNAI. Another week, another worrying disclosure from OpenAI. This time, the company admitted that its agents have been actively evading being caught. The programs did that by turning off internal system meant to flag misbehavior. I guess the human equivalent would be shooting at the CCTV camera while you're trying to rob a bank. So, OpenAI, please install better cameras. According to the Wall Street Journal, it was Elon Musk who ordered restrictions on Ukraine's access to the military version of Starlink last week. Kyiv's forces heavily rely on the network to fend off Russian attacks and conduct their own strikes. The report says Musk discussed the move with President Trump and other officials, but the administration so far has denied involvement. And a more uplifting story from Ukraine, if you will. A Ukrainian-made heavy-lift drone reportedly evacuated a wounded soldier who had been stranded for 491 days. In the world's first operation of this kind, mines, flooding and Russian drones had made a ground rescue impossible. Now, Ukrainian drone technology is redefining what's possible on the battlefield. Poland wants to bring AI assistance into its courts. The country's homegrown model, known as Plum, will face accuracy tests before helping judges with administrative cases. That fine you hoped they wouldn't have time to process? Well, it may come back to haunt you. Every two years, Russians celebrate democracy by meddling in U.S. elections. This year's festivities have an extra feature, AI. Michiel von Brunnenstein reports. Guess who, it's Doc here, or a pretty good impersonator, I'll tell you that. This is not famous actor Christopher Lloyd railing against Democrats, nor is this actress Alison Hannigan doing the same. Both clips use actual footage of the Hollywood actors, but follow it up with an AI-generated voiceover. They are part of Matryoshka, a Russian operation flooding social media with AI-generated disinformation. Now Matryoshka is a 19th century Russian toy that contains a wooden doll within a wooden doll within a wooden doll. It used to be associated with innocence, but not anymore. They use bots, or not bots, accounts that are stolen or purchased from online, older accounts that they've obtained in order to place these stories into the digital ecosystem, usually on X or Blue Sky or Tik Tok. So using AI, a Russian operative would create footage, slap the logo of a trusted media outlet on it, and then, using fake social media accounts, would inflate the number of views to ridiculous levels. I would call it multistage influence operation of like infrastructure that actually is designed to sow doubt and undermine trust and credibility of Western media, fact-checkers, opinion leaders and overall democracy. Common themes of Matryoshka operations usually involve making Western politicians and societies look corrupt and morally bankrupt. The quality of these fakes is usually highly questionable, but Matryoshka is not about making people actually believe actor Eric Roberts went on CNN to rail against vaccines. are talking about it and journalists see it and journalists write about it. And that alone puts these narratives into the ecosystem. It causes us to talk about Russian disinformation and some of these ridiculous narratives that they spin. Wait, are we at TVP World now actually helping the Kremlin here by reporting on it? This is a complicated question I have and conversation I have, honestly, with a lot of journalists. I do think that it's important to talk about uh and important to discuss these campaigns because every now and then rarely one of these matryoshka uh narratives does pick up some viral some virality to it and people to talk about it meanwhile matryoshka has moved on to create narratives about the riots in France for instance uh sites and uh architecture that was attributed previously for Matryoshka was used to amplify narratives that protesters actually are protesting against Macron's militarization policies rather than just poor condition in any educational system. Sometimes they do try to convey some very particular ideas, very particular narratives, but very often this is not a sprint, this is a marathon for them. And my guest today is Mattia Caniglia. He's an affiliate professor at the University of Glasgow, and we will be unpacking Operation Matryoshka with him. Hello, Mattia. Thank you so much for joining me on the show. Thank you for having me. Wonderful. So, Mattia, do you think that this operation can actually define the outcome of the midterms in the U.S.? I think we're far from there. I wouldn't say that this operation has the capacity to directly influence. Of course, there is an element of influence, but I wouldn't say that that is going to be definitive. What is important to say, though, is that we see a consistent and a persistent effort from the Russian, because Matryoshka has been attributed to Russian or pro-Russian entities, to undermine confidence in institutions, media, and sources of information in the context of the midterm elections in the U.S. And they have done a number of things, not just... Sorry, you go. But they have done such operations since, I mean, 2016. They have been interfering in U.S. elections, and now we have artificial intelligence in the game. Does that make a difference in how they operate? It does. It really does, because it makes things easier from their side and faster and force us to play a sort of whack-a-mole game. But this is not the only thing. It's also that with the new Trump administration, a lot of the defenses that were built through the years to fight foreign interference and information manipulation were actually taken down. So we have this situation in which the U.S. system is not as strong as it used to be. And at the same time, artificial intelligence is making things a little bit easier on the side of the attackers. So my next question actually would have been, are governments doing anything to prevent these kinds of operations? For instance, X took down around 30,000 inauthentic accounts. These were not just Russian accounts, but also other state actors that were trying to interfere in the election. But this is just a drop in the ocean. Just to give you a sense, like Doppelganger, which is another information operation linked to Russia, in May 2024, over just the three-day operations, deployed more than 25,000 bots, while X took down 13,000, and these are not just the Russian ones. So you see how the discrepancy is there in terms of numbers. But can we actually measure the impact of Operation Matryoshka or Doppelganger, like you said, before it's too late? We have to be very careful when we talk about measuring impacts, because we cannot read how much there is the metrics of engagement on social media or other numbers that we might get in terms of how many media organizations have been targeted to deduce from that that a certain number of individuals has been actively influenced to vote one way or another. So this is something that we have to be very careful about. But at the same time, the risk is there, and you are completely right. This is also why Operation Matryoshka is often called Operation Overload because the intent is to overload the media organization with fact-checking, polishing their name, making sure that there is no content branded as the content of that media that goes around spreading fake news and so on and so forth. There is also a very interesting case, a specific targeting of specific journalists that are working to expose Russian influence. This is the case of Brandy Zadrozny. It's an important U.S. reporter, and there was a fake report accusing this journalist and other journalists of inventing Russian election interference to create favor for the Democrat candidates in these upcoming elections. Matija, before we finish, I would also like to ask, since these operations are so intricate and they are also using now the speed and the added efficacy of artificial intelligence, If a foreign actor like Russia decides to target one of our democracies, do we even stand a chance of fighting against this kind of operations to protect our democracies? I think we do stand a chance, but we also have to step up our game. The U.S. is living a very precarious situation because of the dismantling of certain institutional and also social civil infrastructure that we're putting in place to protect the country from this kind of interference. I would say that in Europe and in other democratic countries, the efforts are going in the opposite directions. of course like this is a game of cats and mouse so it's it's difficult for us and for how democracy are organized to react and to tackle this threat what we are trying to do in Europe and this is the efforts that come from the European Union but also the national level is to move from the monitoring of the situation into the actually tackling and impose higher cost for this kind of operation. Some of these efforts are working better than others, but this is what we have. And there is no silver bullet for these kind of issues. Absolutely. Thank you so much, Mattia Caniglia, affiliate lecturer at the University of Glasgow for being my guest today on UNAI. It was a pleasure. Thank you so much. And thank you for joining me on this episode of UNAI. But for more news on tech, Follow us on UNAI. I'm Aisha Batten.