Purportedly AI-Generated Video Reportedly Depicted Bulgarian Politician Kostadin Kostadinov Falling During Protest

A purportedly AI-generated video circulated on social media reportedly depicting Bulgarian politician Kostadin Kostadinov falling while jumping a fence during a June 8 protest against adoption of the euro in Bulgaria. Reporting said the clip was generated from a still image taken at the protest and bore a partially obscured Haliuo AI watermark. The video was reposted by multiple accounts and reportedly drew more than one million views.

Risk classification

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

The AI tool was used to generate a political deepfake that was shared on social media, reaching over one million views and misleading public opinion regarding a political figure at a protest.

Additional risk subdomains

• 4.3 Fraud, scams, and targeted manipulation: The deepfake targeted a specific politician, manipulating his likeness to falsely depict him falling in a potentially humiliating manner.

Causal factors

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

The incident was caused by a human user intentionally using a deployed AI video generation tool to create and spread a deceptive deepfake of a politician.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The system is used to generate deepfakes, which are subject to specific transparency and disclosure obligations under the EU AI Act.

AI system and alleged parties

• AI system: Haliuo AI
• AI purpose: Deepfake Video Generation; Image Generation
• Behaviour type: Tool
• Alleged developer: Haliuo AI
• Alleged deployer: Visoko naprezhenie, Mario Stefanov, Budilnikbg.com
• Alleged harmed parties: Kostadin Kostadinov, General public of Bulgaria, General public, Epistemic integrity

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 Minor
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Minor, 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 malicious synthetic content created and spread directly by the incident.

Directly caused: A deepfake video of politician Kostadin Kostadinov falling was created using Haliuo AI.

Indirectly caused: The video was shared on Facebook pages like 'High Tension' and Budilnikbg.com, accumulating over one million views.

Democracy

Reported: The report explicitly describes the spread of political misinformation that could erode democratic norms.

Directly caused: The deepfake falsely depicted a political party leader falling at a protest, potentially manipulating public perception of the politician and the event.

Indirectly caused: The video reached over one million views, with many social media users believing the fabricated event was real, thereby polluting political discourse.

Epistemic

Reported: The report explicitly describes epistemic harm through the fabrication of a video showing a non-existent event.

Directly caused: A fake 6-second video was generated from a still photo to show a politician falling over a fence.

Indirectly caused: Over one million viewers were exposed to this false information, with many believing it to be real as the accompanying text lacked satirical indicators.

People affected

• Occurrences reported: 1
• People reportedly harmed: 1
• People reportedly exposed: 1000000

Potential causes

Management

• Engagement-Driven Moderation: Social media pages prioritized viral reach over factual accuracy.

Technology

• Easy Watermark Evasion: The Haliuo AI watermark was easily covered up using '#' symbols.
• Realistic Image-to-Video Tech: AI tools can generate highly convincing videos from a single static photo.

Data Inputs

• Use of Authentic Source Photo: Using a real news photo as input increased the perceived credibility of video.

Human Factors

• Deliberate Misleading Sharing: Users shared the video as real news rather than satire to deceive others.
• Lack of Audience Verification: Social media users shared and liked the video without checking its source.

Process and Methods

• Inadequate Platform Moderation: Social networks failed to detect and flag the AI-generated video promptly.
• No Automatic AI Detection: Platforms lacked automated tools to identify and label synthetic media.

Regulatory Environment

• Weak Synthetic Media Laws: Lack of legal frameworks to penalize deceptive AI-generated political content.

Information quality

• Classification confidence: High
• Reason for confidence: The fact-checking report is highly detailed, identifying the specific tool used (Haliuo AI), the source image (BTA photo), the exact dates of posts, view counts (over 1 million), and specific visual artifacts proving it is a deepfake.
• Alternative interpretations: The video was posted by a page known for satire, so some might interpret it as harmless political satire rather than malicious disinformation, though the lack of clear labeling led many to believe it was real.

A deepfake video of Bulgarian politician Kostadin Kostadinov was created using Haliuo AI and reached over one million views on Facebook. While it demonstrates the growing accessibility of synthetic media tools for political disinformation, the incident had minor national security impact, with no disruption to critical infrastructure, government functions, or broader societal stability.

• Overall national security impact: Minor
• Scope: Single nation
• Primary target: Bulgaria
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. Represents an ongoing strategic concern regarding the proliferation of synthetic media rather than an active, imminent national crisis.
• Autonomy: Human-controlled. The AI system Haliuo AI was directly controlled by a human user who uploaded the source image and initiated the video generation.
• Novelty: Established threat. The use of generative AI tools to create political deepfakes is an established threat with numerous prior occurrences globally.

Impact by dimension

• Physical security: Negligible. No physical systems, infrastructure, or human safety were threatened by this digital deepfake incident.
• Information security: Minor. AI-generated deepfake of a politician reached over one million views, demonstrating the ease of spreading localized political disinformation, though it was quickly debunked by fact-checkers.
• Sovereignty: Minor. While targeting a political figure, the incident did not disrupt electoral systems, border controls, or core government decision-making processes.
• Economic security: Negligible. No threats to financial systems, critical supply chains, or strategic technologies were observed.
• Societal stability: Minor. The video misled some members of the public regarding a protest, but did not result in large-scale civil unrest, mass surveillance, or systemic human rights violations.