Reported AI-Generated Videos Depicted Former Kosovo President Vjosa Osmani and Other Public Figures in Allegedly False Compromising Scenes

Videos reported as AI-generated depicted former Kosovo President Vjosa Osmani, Lumir Abdixhiku, Edi Rama, Baton Haxhiu, Richard Grenell, and Aleksandar Vučić in fabricated election-period scenes implying territorial-partition talks, an Osmani-Abdixhiku alliance, and opposition coordination against Albin Kurti before Kosovo's June 7 snap elections.

Risk classification

• Primary risk domain: 4 Malicious actors
• Primary risk subdomain: 4.1 Disinformation, surveillance, and influence at scale

The incident involved the coordinated use of AI-generated deepfakes to conduct a disinformation campaign aimed at manipulating public opinion and political processes during an election.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The incident was caused by human actors who intentionally created and distributed AI-generated deepfakes to influence the Kosovo elections.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The incident involves AI-generated content such as deepfakes, which are subject to transparency obligations under the EU AI Act.

AI system and alleged parties

• AI system: Hive, unspecified (Hive)
• AI purpose: Deepfake Video Generation; Deepfake Image Generation
• Behaviour type: Tool
• Alleged developer: Image generator technology developers, Deepfake technology developers
• Alleged deployer: Information manipulation actors in Kosovo, Information manipulation actors, Deepfake creators
• Alleged harmed parties: Voters in Kosovo, Voters, Vjosa Osmani, Social media users, Richard Grenell, Lumir Abdixhiku, General public of Kosovo, General public, Facebook users in Kosovo, Facebook users, Epistemic integrity, Edi Rama, Democratic integrity, Baton Haxhiu, Aleksandar Vučić, Albin Kurti

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Minor, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Minor
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Minor
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: The report explicitly describes the creation and spread of malicious deepfake content.

Directly caused: AI-generated deepfake videos and images were created and posted on Facebook to depict political figures in compromising situations.

Inferred additional harm: It is likely that the content was shared and viewed by thousands of Facebook users before being flagged as fabricated.

Democracy

Reported: The report explicitly describes the dissemination of fabricated videos in the lead-up to snap elections, which threatens democratic norms.

Indirectly caused: The publication of deepfakes during an election campaign seeks to mislead voters and manipulate the democratic process.

Inferred additional harm: Such campaigns can erode public trust in democratic institutions, election integrity, and political leadership.

Epistemic

Reported: The report explicitly describes epistemic harm through the creation of fabricated videos and images.

Directly caused: AI-generated media falsely depicted political figures discussing the 'division of territories' and dancing at rallies.

Indirectly caused: The spread of these deepfakes on Facebook contributed to the dissemination of false narratives and misinformation.

Inferred additional harm: Widespread exposure to such deepfakes can degrade the shared information ecosystem and make it harder for citizens to distinguish real events from fabricated ones.

People affected

• Occurrences reported: 1
• People reportedly harmed: 6
• People reportedly exposed: 6

Potential causes

Management

• Inadequate Platform Oversight: Facebook management failed to allocate resources to prevent regional deepfakes.

Technology

• Generative AI Deepfakes: AI tools allowed highly realistic fabrication of political figures.
• Lack of Automated Prevention: Social media platforms lacked active AI filters to block deepfakes at upload.

Data Inputs

• Exploitation of Public Media: Real images of politicians were used to train and generate fabricated videos.

Human Factors

• Malicious Political Intent: Bad actors created deepfakes to influence voters before early elections.
• Public Susceptibility: Users share sensational fabricated content without verifying its authenticity.

Process and Methods

• Reactive Fact-Checking: Verification by Krypometer occurred only after the videos gained traction.
• Inadequate Content Screening: Facebook processes failed to flag political deepfakes during upload.

Regulatory Environment

• Weak Digital Campaign Laws: Lack of strict legal frameworks targeting political deepfakes during elections.

Information quality

• Classification confidence: High
• Reason for confidence: The report clearly details the nature of the deepfakes, the political figures involved, the platform used for dissemination (Facebook), and the fact-checking process using Hive. There is little ambiguity about the events described.
• Ambiguities identified: The specific AI generator used to create the deepfakes is not identified, only the detection tool (Hive) is named.

AI-generated deepfakes targeted Kosovo political figures, including President Vjosa Osmani, during snap elections to falsely imply secret territorial negotiations. While representing no kinetic threat, the campaign highlights the evolving risk of AI-enabled information warfare targeting democratic processes and sovereignty.

• Overall national security impact: Substantial
• Scope: Multiple nations
• Primary target: Kosovo
• Other affected: Albania, United States
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Near-term. The threat was active during an ongoing election campaign, requiring near-term monitoring and fact-checking response.
• Autonomy: Human-controlled. The AI system functioned strictly as a generation tool under the direct control of human creators who designed and distributed the media.
• Novelty: Evolved capability. Represents an evolved capability where deepfakes are increasingly sophisticated and targeted at key political figures during critical democratic events.

Impact by dimension

• Physical security: Negligible. No physical threats, kinetic attacks, or critical infrastructure disruptions were reported in connection with this incident.
• Information security: Substantial. AI-generated deepfakes were used in a coordinated disinformation campaign targeting prominent political figures to manipulate public opinion during a sensitive election.
• Sovereignty: Substantial. The campaign targeted key leaders during snap elections, directly threatening democratic processes and the integrity of government formation in Kosovo.
• Economic security: Negligible. No significant threats to financial systems, strategic industries, or technological competitive advantages were reported.
• Societal stability: Minor. The disinformation had the potential to erode public trust in democratic institutions and political leadership, though large-scale unrest was not reported.