Why develop your political movement's ideology when you can generate it in an instant with artificial intelligence? Attention spans are short anyway. Complex narratives are redundant, aren't they? Feeding off stereotypical visual references to an ideological past is enough, right?
AI-generated aesthetics are becoming inseparable from a globalized form of contemporary extremist narratives leveraged by far-right political parties. The wording used in prompts for large language models (LLMs) to create idealized digital images transforms the perceived identity of the nationalist ‘in-group' – and impacts reality. In Romania and Moldova, for example, the scale of AI-bot-flooded far-right election campaigns set alarming precedents in 2025. Similarly, AI enhances the currently effective German, far-right AfD party's rhetoric.
Grassroots Slovak nationalism
Small nationalist groups are equally active in creating their idealized identity, adding to the morass of AI-generated imagery to increase their supporters' perceptions of belonging. While also highly generalized, their imagery taps into localized historical references.
The Slovak, pro-Russian bikers club Brat za Brata (Brother for Brother), for example, regularly posts AI-generated images on Facebook such as pictures of young, white, blond men and women dressed in traditional costume, of muscular ‘medieval' Slavic brotherhood, and of anthropomorphized wild animals representing Russia, Slovakia and Czechia standing shoulder to shoulder. The images are often set in approximations of the great outdoors, dramatically lit for heroic effect. Contemporary propaganda aesthetics are a layered jumble of elements: generic depictions of nature and stylized people meet nationalist symbolism within fantasy settings reminiscent of CGI movies for the whole family. Historical stereotypes do not disappear. They are heavily referenced within new output. AI just makes them easier to generate quickly and repetitively.
Images from Brat za Brata Facebook page.
In contrast, actual pictures of Brat za Brata show a group of middle-aged men posing together in urban locations, wearing black bomber jackets emblazoned in Cyrillic. The only constant between the two sets of imagery is that the bikers are flying the Russian flag in both.
While claiming to be apolitical, Brat za Brata, which organizes Soviet memorialization events, have been using their online platform to actively promote a pro-Russian agenda since the full-scale invasion of Ukraine. The movement's leader, Matúš Alexa has met Sergei Naryshkin, the head of the SVR, Russia's foreign intelligence service. Members regularly attend events at the Russian embassy in Bratislava, boasting of their relationship with former ambassador Igor Bratchikov. In March 2025, the group delivered a questionable petition to the Slovak president calling for sanctions against Russia to be withdrawn.
Image from Brat za Brata Facebook page.
The soft power of a cartoon bear
While saturated in seemingly wholesome innocence, Brat za Brata's representations of a ‘true' Slavic identity take their place among an onslaught of other AI-generated far-right imagery that advocates for strict nativism. Russian state-owned media, aficionados of AI-generated propaganda, regularly hit out at the legitimacy of international democratic institutions, especially in relation to Ukrainian membership. In a series of provocative images posted by RIA Novosti, rats with demonic red eyes and bared fangs clammer within a box labelled ‘NATO' while others wearing the Ukrainian flag follow the pied piper or stage a peasants' revolt under an arch of EU stars.
Although no single social media post might be able to shift public opinion, collectively they can be highly instrumental. The real potential of AI-generated imagery is in its openness to the incremental radicalization and polarization of identity and myth-building. ‘If you don't construct a grand narrative – saying something here and there but never allowing the full, dangerous picture to come together –' it's ‘difficult to apply the laws we commonly have in Europe regarding hate speech …, because everything remains merely suggestive,' says philosopher of language Tomáš KoblÞek.
AI-generated images used by RIA Novosti.
Even less-explicitly pro-Russian-narrative content is therefore also causing controversy. Recently, the widely available Netflix children's show Masha and the Bear, for example, faced a backlash in Ukraine and the UK for likening its ‘gentle-giant' bear to Russia. Making the information space friendlier to broadly pro-Russian narratives can be an exercise in soft power. In avoiding strictly political, polarizing opinions, soft power complements traditional diplomacy, communicating via civil society channels.
Some propaganda imagery is intentional nonconfrontational, whether it takes the form of the bear from a kid's cartoon or Brat za Brata's ‘Disney-inspired' Facebook posts, because inoffensive representations circulate freely. The large-scale production of AI-generated far-right content, saturating the online environment, is increasingly becoming mainstream. Whereas the use of far-right extremist symbolism is frowned upon or is even illegal in certain countries, newly generated content based on ideological backgrounds is perceived as mild when not depicting violence directly. Nationalist groups and political parties hide racist, homophobic, patriarchal meanings in softened visual narratives to appear, at least on the surface, normative and surpass digital platform restrictions.
When the integrity of political reality is broken down piece by piece, starting in small communities through seemingly innocent channels, the way is paved for similar tactics to influence national elections without scrutiny. Political actors leverage the use of AI beyond content generation, gathering data to create detailed psychological profiles of the electorate for highly personalized messaging. Covert operations producing large amounts of AI-generated content also allow politically motivated creators to see which topics and approaches resonate best. Occasional fear mongering is considered fair play, especially prior to voting, just to remind everyone of who supposedly public enemy number one is. But satire and other uses of humour are also proving to be particularly effective.
How amusing are deepfakes?
Caricatures are prominent in generative AI, which enables the production of infinite variations of people set up in imaginary scenarios. The production of deepfakes based on, for example, images of real politicians' faces has sparked debate about the ethical boundaries of such visual manipulation.
During the 2023 election moratorium – the legally prescribed period for campaigning in Slovakia – a deepfake scandal hit the internet: a recording faking a phone call between Michal Å imeÄka, leader of the Progressive Slovakia party, and journalist Monika Tódova initially spread through Telegram. What was surprising just a few years ago, however, has since become an everyday part of the social media ecosystem. More recently, a deepfake video of Slovakia's current coalition politicians and an oligarch supposedly in jail made by an Instagram user has also gone viral – in part due to it being shared by opposition politicians.
While predisposed to deriding their opponents, following these examples and new EU legislation, which came into effect at the beginning of August 2026, Slovakian politicians are tackling deepfake videos. A full ban plus punishment ranging from a fine to a potential prison sentence for up to two years for offenders are in the offing. Experts have meanwhile raised concerns about assessing accountability: not only questioning whether those who prompt AI or the models themselves are at most fault but also recognizing the difficulty of finding the actual creator of harmful content. In some cases, it is clear who posted an image, like on the Brat za Brata Facebook page. But many accounts are anonymous and AI-generated content is often spread through bot networks, removing culpability a stage further. In Europe, content distributed on large online platforms is the responsibility of the Digitial Services Act, but its operation is more often than not reactive, focusing on social media platform accountability rather than on regulating the AI tools used to generate misleading content.
Top row, from left to right: Matúš Šutaj-Eštok (minister of interior), Robert Fico (prime minister).
Bottom row, from left to right: Peter Pellegrini (president), Pellegrini and Juraj Blanár (minister of foreign affairs), Zuzana Čaputová (former president), Čaputová and Ľuboš Blaha (Slovak MEP), Peter Pellegrini. Image via madeinslovakai Instagram account.
Proposed legislation for safeguarding digital identity may aim to protect people from having their face misused for deepfake content, but it addresses only part of the issue. It doesn't begin to tackle the overflow of AI-generated imagery across the political spectrum in the first instance. Neither does it confront ‘plausible deniability' invoked to evade responsibility for offensive behaviour: ‘If someone accuses you of, say, racism or homophobia, you can simply claim it was a joke,' says KoblÞek. And it doesn't ask whether generative AI would be as effective if it weren't for existing prejudices.Â
Reviving reactionary modernism
‘If you look at how the German weekly tabloid … Der Stürmer depicted Jews in the 1930s, it is very similar to how the AfD depicts migrants today using AI-generated images,' comments researcher of political iconography Jacopo Galimberti. Nazi images of idealized Aryan families could be similarly compared to those produced by Brat za Brata. Stereotyping is a common nationalist tool that is especially effective in visual form, exaggerating the differences between the ‘ideal' and ‘the other'.
Much Nazi propaganda relied on figurative representations of subjects such as the traditional family, employing recognizable depictions devoid of visual complexity and yet bursting with civic pride. Machines were equally glorified. The Nazi's enthusiasm for technological progress was focused on stopping the supposed decline of Western culture and calling for a form of spiritual awakening. Post-war, Jeffrey Herf termed this vision ‘reactionary modernism', a movement ready to paradoxically apply technological development in a bid to return to traditional nationalist myths and ideas.
Cover of the ‘Neues Volk' monthly magazine calendar from 1938. Image via Wikimedia Commons.
Recent developments in digital technology, including AI tools and their uses, show parallels to twentieth-century reactionary modernism. ‘Technofascism' describes the ideology behind technological expansion over the democratic infrastructure of the state, empowering Big Tech companies into government decision-making roles, under the direction of the likes of Peter Thiel and Elon Musk. The aesthetics of Tesla's cybertrucks and ‘trad wives' come worryingly close to the Nazi regime's obsessions with military technology and racially pure families.
What has changed, however, is the participatory aspect of generative AI, which makes a content producer out of every consumer. Nazi propaganda was orchestrated for the masses rather than by the masses. Joseph Goebbels saw film as the main medium for communicating National-Socialist ideology, utilizing its entertainment factor to further wartime propaganda. Movies from the period such as popular comedies, while seemingly telling apolitical stories, portray racist policies. Film's influence remained so key to the regime that even a few months before Germany's capitulation, people were still being ‘drafted' as extras for the National Socialist's last-minute productions.
Today's generative AI goes a step further than engaging civilians as extras on the big screen. It offers consumers full participation and customization during content creation, giving them control over their visualizations on the small screen. Being free to make, AI-generated images feed into grassroots forms of propaganda. Now, provocative content can go viral, assisted by automated bot networks boosting posts on users' feeds, disguised as clickbait. Members of Brat za Brata from Revúca, a town of 10,000 people in central Slovakia, don't need a producer or a local audience. In a couple of clicks, they can visualize all of their ideas on identity and power structures, and communicate them to a network of users who readily consume nationalist messages.
The enshittifying reality of platform realism
Relatively new yet already pervasive, AI-generated online content, produced in a bright and colourful pallet with a recognizable visual characteristic, has already earned itself a scholarly term: ‘platform realism'. Coined by Jacob Birken, and further developed by Roland Meyer, the concept identifies generative AI's characteristics of legibility, plausibility and structurally nostalgic relationships to past media. Meyer highlights how the very origin of generative AI products separate them from reality as we know it, starting with how LLMs process the prompts they receive. As Meyer explains: ‘despite the popularity of the term “prompt engineeringâ€, their function is not that of an exact program to be executed but that of fuzzy parameters narrowing down possible results and steering the model toward a specific region within a seemingly infinite statistical space of possibilities.' In the end, the user has to try multiple times to get an image that they like, which is inherently a wasteful process both in terms of resources and littering the online space with visual waste, nostalgia and clickbait.
In attempting to analyse this new reality, researchers across many fields, from philosophy and sociology to studies of processing and information systems, build on Jean Baudrillard[i]'s concept of simulacra and hyperreality. Baudillard's theory, which describes how representations can feel more real than reality, takes Disneyland as an illustration of how hyperreality replaces the reality beyond the theme park. He argues that we are even past that point: the ‘real world' is now full of simulacra, making it almost impossible to find ‘real' representations anymore.
The new-level simulacra model is an autonomously generated reality, approximated from our knowledge archives. The creation of art, or in this case, the creation of propaganda, previously requiring human imagination, has mutated into a reduced series of prompts for AI models. And the representations that were supposed to convince us of the real world that surrounds us have been replaced by fully artificial representations.
In line with philosopher Marshall McLuhan's media theory of technology as ‘an extension of consciousness', AI appears to provide members of Brat za Brata with their dream reality. The group's AI-generated imagery may well even adjust their sense of self – assuming that these aesthetic choices shape their perceptions of reality. As philosophers Philippe Lacoue-Labarthe and Jean-Luc Nancy discuss in their 1990 essay ‘The Nazi Myth', the ‘mythical power is the power of the dream, of the projection of an image with which one identifies.' While the tools used to portray and communicate identity have changed dramatically, the old myth of Slavic unity carries on, growing in endless exaggeration thanks to groups such as Brat za Brata. But the creation of imagery with seemingly infinite possibilities comes with distinct caveats.
‘Machine learning works by correlation, by analogy, in a way more like alchemy than cause-and-effect-based science,' says digital media artist Hito Steyerl. ‘There is a proximity between magical thinking and elements of machine learning. But this magical vibe is blatantly pushed by large corporations just to flog their products; they're not magical at all, but rather enshittifying.'
Strategic pollution of LLMs
LLMs learn from the largely Western-dominated virtual archives we provide them with. Once patterns have been recognized, they can be reproduced on a mass scale, in infinite variations. The prevalent information is therefore biased.
‘When you stop to think about the issue of responsibility, there is often a tendency to shift the focus onto the AI tools themselves. We talk about what AI has brought us, whether AI denies the Holocaust or not. However, the responsibility actually lies with the creators of the specific content or the providers of the AI model – and these responsibilities can vary,' says KoblÞek.
Technology moves faster than regulation. The rate and complexity of LLM development is such that it has already become impossible to keep up with whether the models are being trained to spread a certain rhetoric on purpose or if the people in charge have failed to prevent these biases. The online space is being shared between regular users and inauthentic websites mass-producing content whose target audience is other bots for future content production. And, as KoblÞek comments, far-right groups are keen for their material to become part of the LLM landscape: ‘Extremist groups actively work to ensure this content is available online – even if it doesn't garner many views – simply so it is there as a potential source for AI training.'
Given the way generative AI functions, it also tends to reinforce the sexist and racist imagery, already existing within the virtual archives feeding LLMs, which far-right actors are seeking. ‘It thus allows the right to conjure up a nostalgic, timeless past that never existed outside of the images used to train the software,' says Meyer. The reciprocal mechanism of supply and demand is in place. Add a format that is easy to share online, and a propaganda machine is born.
In some cases, we see politicians with other political leanings also jump on social media trends, generating their own content that mocks their opponents. However, ‘liberals, progressives and leftists, by contrast, have largely been reluctant to publicly use AI – not only because of the models' well-known biases but also because of the massive exploitation of human labour and creativity, as well as of natural resources involved in training AI models. For the right, then, embracing recognizable AI or “slop†aesthetics is also a way of triggering and trolling their opponents,' adds Meyer.
Soon we may have discussions not only about generating new images but also about protecting the integrity of the past, which, in an Orwellian way, can easily be altered. The US government has already started by erasing images of women, LGBTQIA+ people and people of colour from its military websites. Meyer warns of ‘digital forgetting' – in other words, we shouldn't be so sure of the ‘total availability of the past'. ‘This not only breeds a pervasive distrust in visual evidence of any kind but also produces misleading expectations about what historical documents should look like,' says Meyer. When we erase the originals from our archives, they may well only remain as shadows in the training material of LLMs.
Museums and public institutions sometimes embrace generative AI themselves, to fill in historical gaps and bring the past into life. According to Meyer, ‘such speculative uses are often misleading. They tend to reproduce existing clichés about the past and how it should be represented, rather than opening up a new view of history.'
Institutions are faced with keeping ideas of truth alive, developing public awareness and helping define codes of practice on AI use. Regulating at the speed of developing technology will be a constant challenge.
Meanwhile, AI-generated content from groups like Brat za Brata that can keep being produced ad infinitum may well reach a point of saturation, eventually only persuading the persuaded or simply tiring its viewers out – as with other forms of propaganda before it.
Â
This article is published with the support of the ZEIT-Stiftung Ebelin und Gerd Bucerius.
