This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A report detailing the proliferation of AI-generated deepfake videos in Kosovo, specifically targeting political figures like Dimal Basha, Ramush Haradinaj, and Albin Kurti. These videos are being used to spread disinformation on platforms like Facebook and TikTok, with fact-checkers identifying hundreds of instances of manipulated content. Experts warn that this trend threatens democratic processes, deepens social polarization, and exploits low media literacy in the region.
Minutes after Kosovo MPs elected Vetëvendosje's Dimal Basha as Parliament Speaker, a purportedly AI-generated Facebook video used his likeness to claim he was the father of Kosovo's Ashkali community, a Roma-related minority group that has faced long-running discrimination and political marginalization. The video reportedly received more than 80,000 views in under eight hours.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to conduct large-scale disinformation campaigns, malicious surveillance, or targeted and sophisticated automated censorship and propaganda, with the aim of manipulating political processes, public opinion, and behavior.
Also classified as
The deepfakes include degrading portrayals of political leaders, contributing to political polarization and hate speech.
The rapid spread of unverified AI content on Facebook and TikTok is polluting the local media ecosystem and eroding public trust.
The incident involves coordinated and at-scale dissemination of AI-generated deepfakes targeting political figures to influence public opinion and democratic processes in Kosovo.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident was caused by human actors intentionally using deployed AI generation tools to create and spread deepfakes to manipulate public opinion.
Added categories peak at 3: Substantial (Epistemic Harm, Indirect)
The coordinated spread of AI-generated deepfakes targeting Kosovo's political leadership poses a substantial threat to information security, democratic processes, and societal stability. Exploiting low media literacy, these campaign efforts polarize the public and threaten electoral integrity, potentially amplified by external adversarial networks.
The AI systems are used as tools by human creators to perform the specific function of generating synthetic video and audio content based on direct inputs, without exhibiting independent goal-directed behavior or autonomy.
Limited Risk: The report describes the generation of deepfakes, which under the EU AI Act are subject to transparency obligations to ensure users are informed that they are interacting with AI-generated content.
The AI systems were used to generate manipulated videos and fabricated audio recordings (deepfakes) of political figures.
The reports provide clear, consistent accounts of the deepfake phenomenon in Kosovo, backed by specific statistics from certified fact-checking organizations (Krypometer, BIRN Kosovo) and quotes from local experts and political representatives. The role of AI as a tool for generating these deepfakes is explicitly stated.
None
Technology
Rapid AI Generation Tech
Rapid development of AI fuels easy creation of manipulated video and audio.
Highly Realistic Deepfakes
AI-generated content looks so real that audiences fall prey to propaganda.
Platform Removal Failure
Social platforms fail to provide methods to remove AI-generated content.
Data Inputs
Fabricated Audio and Video
Manipulated biometric data is used to synthesize fake political statements.
Unlabeled AI Content
AI materials are shared by users without any clarification of being fake.
Foreign Disinformation Feeds
Kremlin-linked networks feed disinformation to multiply hate speech.
Human Factors
Low Media and Tech Literacy
Low technological literacy leads to high vulnerability to AI deepfakes.
Vulnerability of Older Adults
Older generations take deepfakes seriously, reinforcing political beliefs.
Unchecked Sharing by Users
Regular users share fake content thousands of times without verification.
Process and Methods
Delayed Fact-Checking
By the time fact-checkers step in, the viral damage is already done.
Lack of School Curriculum
Public school curriculum lacks media and artificial intelligence literacy.
No Dedicated Verification
Influential media outlets lack dedicated teams to verify AI content.
Management
Platform Governance Failure
Social media companies fail to manage and remove harmful deepfakes.
Regulatory Environment
No State Institutional Response
State institutions fail to provide a serious response to AI manipulation.
AI system involved
An AI system was involved, but the reports do not identify it.
Identified by automated extraction; can differ from the alleged developers above.
