By Noura Kaddaoui
A new political weapon is spreading through social media feeds, often disguised as entertainment. Sometimes it looks like a Lego animation, a rap video, a fake influencer, a battlefield clip, or a meme. It is cheap to make, fast to publish, and easy to tailor to different audiences. Researchers call it slopaganda: AI-generated propaganda content designed to manipulate political beliefs, emotions, and attention.
The word was coined by researchers Dr. Michal Klincewicz, Dr. Mark Alfano, and Dr. Amir Ebrahimi Fard. They define it as unwanted AI-generated content spread to manipulate beliefs for political ends. Their core argument states that generative AI changes propaganda through speed, scale, and personalisation. “Generative AI enables a whole new quality, strategy, and tactics of mass persuasion that were never available before,” says Dr. Klincewicz.
The same AI tools used to flood feeds with geopolitical propaganda are already appearing in extremist and hate-driven campaigns. Investigations and research have documented AI voice cloning by extremist and terrorist groups, alongside AI-generated anti-Muslim, anti-Christian, and anti-migrant narratives.
Cheap, fast and personalised
Traditional propaganda required resources. Posters had to be printed, broadcasts produced, and troll farms needed workers. AI removes much of that friction. “Almost instantaneously, you can have a video, a news story, or an image that can be used to fill some informational niche that is not been filled yet,” says Dr. Klincewicz.
Slopaganda also allows a level of personalisation that traditional propaganda never had. Unlike a leaflet or television programme, AI generated propaganda can be tailored around user characteristics. “For example, a propaganda poster about Napoleon would once have shown the same message to everyone. With generative AI, the poster can change depending on whether the audience is Catholic, Orthodox, Italian speaking, Greek speaking or motivated by different fears,” says Klincewicz. AI-generated campaigns can repackage one version of the same prejudice for different audiences. For example, one anti-immigrant message can be framed as economic anxiety, cultural loss, religious threat, or security panic, depending on the receiver.
Journalism and Political Communications Professor Vian Bakir, from Bangor University, points to AI-generated rapper Danny Bones as a warning sign. “Bones is built and managed by the Node Project and was commissioned by Advance UK, a far-right party, to produce content about immigration, national identity, and cultural decline.” Bakir warns that AI can “generate large scale propaganda or disinformation campaigns” by flooding platforms with deepfakes or creating influencers “to say bad things about marginalised groups”. Dr. Klincewicz makes a similar point about Roma people, LGBTQ communities, women, and other minorities, warning that fake videos can repeat familiar stereotypes “at scale”.
Scrollable propaganda
One of the most recent and well-known cases of slopaganda is the pro Iran Lego style AI video campaign. The videos show US President Donald Trump, Israel’s Prime Minister Benjamin Netanyahu, Iranian soldiers, and war scenes as Lego style figures, often mixed with rap music, jokes, and pop culture references. The videos were boosted by Iranian state media and Iranian government accounts. The Lego aesthetic makes the videos easy to watch and harder to process as formal war propaganda. In its main mechanism lies humour that carries the political message where “the joke is not the message, but the delivery system”. These clips have reached audiences who were not actively following the war.
According to Klincewicz, the danger is not just limited to whether a clip is real or not, because the artificiality in these kind of videos can be obvious. “Nobody that looks at videos that are obviously fiction believes them to be true,” he says. The effect comes from repetition and emotion. Slopaganda can present the same ideas across videos, images and texts until it “reinforces associations we are not even conscious of believing,” Dr. Klincewicz says
Professor Bakir points to a more subtle form of political messaging. “Cheap methods such as deepfakes and AI influencers can generate large scale propaganda or disinformation campaigns against marginalised groups”, she says, while making the message look like ordinary content. Bakir warns that AI can also create “a false consensus” where fake accounts make hostility toward a group look widespread.
Fake content, real damage
RNW journalist and information integrity manager Giovana Fleck says targeted AI abuse can push people out of public debate.
“It is easy to miss the groups that go silent because they are afraid of being targeted,” she says. “Self-censorship is still censorship.” Fleck refers to deepfake abuse and non-consensual sexual imagery targeting women. “You do not need a huge amount of it out there for victims to silence themselves,” she says. If women, queer people, migrants, racialised groups and other targeted communities leave public conversation, the information ecosystem becomes narrower. She also stresses that AI systems carry bias. “There are human decisions that dictate how content is shared, created, and used inside different AI systems.”
Klincewicz, Alfano, and Ebrahimi Fard state that corrected misinformation can still leave a trace in memory and continue shaping how people interpret later information. “These representations do not disappear like the text we delete in a document. A trace of the prior representation typically remains in the brain, influencing how new information is integrated with prior knowledge.” That means a fake image does not lose all power once it is debunked. If, for example, AI content repeatedly shows Muslims as violent, migrants as threats, or queer people as dangerous, debunking one image does not erase the wider impression.
Regulation is still catching up
The EU AI Act and the Digital Services Act (DSA) address different parts of the problem. The AI Act requires transparency around AI-generated or manipulated content, including deepfakes. The DSA puts duties on large platforms to assess and reduce systemic risks such as disinformation and threats to public debate. Dr. Klincewicz calls both “a first good step,” but says Europe still needs “much more focus on generative AI content”. He calls for mandatory watermarking, fines when platforms circulate unlabelled AI material, and accountability for large tech companies. The Center for the Study of Organized Hate also recommends political AI content to be disclosed, synthetic disinformation to carry penalties, and AI developers to publish regular risk assessments.
Professor Bakir points out that current policy is still too focused on single pieces of content, while AI campaigns can now simulate public opinion. “Current policies are not up-to-speed on AI-generated mechanisms for orchestrating false consensus online. Deepfake labels also have limits. “Under article 50 of the EU AI Act, deepfakes must be disclosed as manipulated or artificially generated. However, such transparency or labels can easily be lost as images and videos are shared online,” Bakir says. That confusion can feed “widespread distrust,” she says, because “we have to be able to trust what is real and factual”.
Journalist Fleck is sceptical of leaving platforms to police themselves. “Technology advances faster than regulation can keep up with,” she says. “When regulation tries to catch up, it is done in a sloppy way that opens the door for possible harms.” Fleck finds that affected communities should be part of the policy process from the start, because the risks are different for migrants, women, queer people, activists, and other targeted groups. “There is no one-size-fits-all solution.”
The European Federation of Journalists states that Europe also needs a stronger media ecosystem, such as independent journalism, media literacy, and limits on platform power. Regulation can label or remove some synthetic content, but it cannot rebuild trust on its own.
Expose the manipulation, not the stereotype
Dr. Klincewicz warns against reducing slopaganda to meme culture. “This is intentional and well designed by people who know what they are doing,” he says. “If these images are made to be cute and funny, they are hiding the atrocity underneath it.” According to the researcher, “war clips styled as Lego memes or fantasy images of Gaza as a luxury Riviera may look absurd, but they depict real conflicts with real consequences”.
Professor Bakir says audiences should be made aware when AI generated propaganda exists, especially in cases such as Danny Bones, that “are sponsored by a political party to encourage Islamophobia”. She says that harmful AI propaganda should be labelled as AI generated and, where possible, accompanied by context on who funded it. At the same time, she urges journalists to avoid panic-driven headlines or broad warnings that make people doubt everything they see.
Fleck says newsrooms also need to look at their own role in the AI ecosystem. They are under pressure, she says, but they still have to understand what happens to the data, images, and stories they publish. “We need to educate ourselves about the data that we share with platforms, especially when it comes to sensitive data from our investigations and sources,” she says. A photo from a refugee camp or a testimony from a woman targeted by harassment can be reused by AI systems in ways the source never agreed to. “We have a duty of care when it comes to the people that we portray.”
Research suggests that corrections and warnings can reduce false beliefs, but they can also make people more skeptical of information in general. Slopaganda reporting also has to show where the content came from, who it is meant to influence, and which prejudice it is trying to reinforce. When AI-generated content is built around racist, sexist, or xenophobic stereotypes, reporting needs to expose the manipulation without reproducing the harmful framing.
The views and opinions expressed in this article are solely those of the author and do not reflect the official policy or position of the Media Diversity Institute. Any questions or comments should be addressed to the editor at [email protected] of the Media Diversity Institute (MDI).
