A Polish journalist’s three-day investigation on TikTok revealed a surge of AI-generated propaganda during the Hungarian election campaign, proving surprisingly effective.
AI Propaganda on TikTok
The Hungarian election campaign heavily utilized TikTok, with deepfakes and AI creating realistic propaganda difficult to distinguish from truth. A journalist spent three days assessing how easily she could be misled, documenting the results of this campaign experiment.
One example involved American boxer and YouTuber Jake Paul filming a group of elderly Fidesz voters singing praises to Viktor Orbán, while symbolically bowing to a sack of potatoes – representing electoral bribery and Hungarian poverty. The video garnered 160,000 views.
Supporting Tisza with AI
Another video, generated by AI, showed Jake Paul supporting the Tisza party, shouting “Tisza forward!” surrounded by orange-clad seniors (orange being associated with the competing Fidesz party) who aggressively confront him. This video received nearly half a million views.
Kremlin Involvement and Pro-Government Networks
These deepfakes largely evaded analysts focused on pro-government accounts discrediting opposition candidate Péter Magyar and his supporters – numbering in the hundreds. Independent portal “VSquare” and the “Financial Times” reported on the Kremlin’s involvement in operating troll farms producing AI content for the Hungarian campaign.
Synthetic Propaganda and Voter Perception
During her three days on Hungarian TikTok, the journalist found most videos were AI-generated propaganda from troll accounts. These depicted Magyar as a puppet of Brussels, or portrayed him negatively through fabricated stories about his lifestyle.
One animation featured national hero János Hunyadi warning that Magyar would lead Hungary into the war in Ukraine.
Tisza’s Reach vs. Fidesz’s Resources
Péter Krekó, director of the Political Capital think tank and coordinator of research on disinformation at the Hungarian Digital Media Observatory, stated that Tisza was outperforming Fidesz in reach. According to his analysis, the opposition generated twice the engagement of Viktor Orbán’s party.
Organic Reach and Deepfake Recognition
Much of Tisza’s success stemmed from fan-made content and Péter Magyar’s official account, rather than bot farms or AI-generated content. While Fidesz utilized AI, their content – such as animations featuring Magyar or historical figures – was often easily identifiable as artificial.
However, accounts like the Jake Paul deepfake and @pannavarga.96 proved deceptively realistic, successfully misleading the journalist.
The Power of Confirmation Bias
One TikTok post from @pannavarga.96 accused the government of selling land to Chinese factories and poisoning the water supply, labeling it as treason. The journalist admitted to spending fifteen minutes looping the video, initially unable to determine if the speaker was real.
Experts explain that advanced AI tools use motion tracking and face-swapping to create highly realistic content. Even weak deepfakes can be effective if they confirm existing beliefs, exploiting confirmation bias.
Evolving Technology and Future Threats
Dobromir Ciaś of Edge NPD emphasized that the technology is rapidly improving, and by 2026, AI will be able to replicate reality and mimic known individuals with greater fidelity. This poses a significant threat to the 2027 Polish elections.
The journalist concluded that even unsuccessful deepfakes can be effective, and credible, realistic imitations are even more dangerous. She noted that she, as a campaign expert and social media native, was herself susceptible to deception.