Purported Deepfake Targets Olena Zelenska in Alleged Russian Disinformation Campaign

A purported deepfake video falsely suggesting that Olena Zelenska, wife of Ukrainian President Volodymyr Zelenskyy, purchased a luxury car, reportedly circulated widely online. The video is reportedly part of a Russian-linked disinformation campaign aimed at undermining Ukraine and its supporters.

Risk classification

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

The incident involves a coordinated Russian-linked network using AI to conduct a large-scale disinformation campaign aimed at manipulating public opinion and political processes in the US and UK.

Additional risk subdomains

• 3.1 False or misleading information: The campaign generated highly specific false narratives, including a fabricated invoice and a deepfake video, to mislead the public.

Causal factors

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

The disinformation campaign was intentionally orchestrated by human actors (John Mark Dougan and Russian networks) using deployed AI systems to generate and spread fake news.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The incident involves 'AI-generated content such as deepfakes' and chatbots (ChatGPT used to generate articles), which fall under Limited Risk requiring transparency obligations so users know they are interacting with AI or viewing synthetic media.

AI system and alleged parties

• AI system: unspecified
• AI purpose: Deepfake Video Generation; Social Media Content Generation
• Behaviour type: Tool
• Alleged developer: Unknown voice cloning technology developers, Unknown deepfake technology developers
• Alleged deployer: Verite Cachee France, Russian-linked disinformation network, Pro-Russian influencers, Pro-Kremlin actors, Misinformation spreaders, Disinformation spreaders
• Alleged harmed parties: Volodymyr Zelenskyy, Truth, Relations between the United States and Ukraine, Olena Zelenska, National security and intelligence stakeholders, Government of Ukraine, General public of Ukraine, General public of the United States, General public of the European Union, Epistemic integrity, Bugatti

Harm severity

Highest direct severity in any category: Severe. 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 Severe, indirect Severe
• 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 Severe, indirect Severe
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: The report explicitly describes toxic and malicious content created and spread directly by the incident.

Directly caused: The AI-generated deepfake video and ChatGPT-generated articles reached over 18 to 20 million views across platforms like X, Telegram, and TikTok within 24 hours.

Indirectly caused: The content was amplified by pro-Russian influencers and state media, spreading the malicious narrative globally.

Inferred additional harm: Further malicious content was likely generated across the network of 167+ fake websites operated by the same actor to sustain the campaign.

Democracy

Reported: The report explicitly describes erosion of democratic norms caused directly and indirectly by the incident.

Directly caused: The campaign targeted US and UK voters on the eve of elections to influence their voting behavior regarding aid to Ukraine.

Indirectly caused: The disinformation was cited by US members of Congress, such as JD Vance, potentially influencing policy debates and democratic decision-making.

Inferred additional harm: The ongoing operation of over 160 fake news sites likely continues to erode trust in democratic institutions and media ecosystems.

Epistemic

Reported: The report explicitly describes epistemic harm caused directly and indirectly by the incident.

Directly caused: The network fabricated a fake purchase invoice and a deepfake video of a non-existent Bugatti employee to launder a false narrative.

Indirectly caused: The fake news was indexed as the top result on Google Search via MSN, misleading users who searched for the topic.

Inferred additional harm: The scale of the AI-generated spam farm contributes to a broader erosion of trust in online information and search engine reliability.

People affected

• Occurrences reported: 1
• People reportedly harmed: 1
• People reportedly exposed: 20000000

Potential causes

Management

• Lack of Aggregator Oversight: MSN lacked proper vetting processes for source sites in its news feed.
• No Editorial Standards: The copy-site network intentionally omitted editorial safety reviews.

Technology

• AI Deepfake Video Generation: Deepfake video of fake Bugatti employee created to deceive viewers.
• LLM Mass Article Generation: ChatGPT used to rapidly generate thousands of politically biased articles.
• SEO Manipulation Algorithms: Digital strategies pushed AI-generated fake news to top search results.

Data Inputs

• Scraped Media Content: Real media articles scraped to provide context for AI-rewritten propaganda.
• Biased System Prompts: Instructions told LLMs to favor Russia and Republicans while attacking allies.
• Stolen Identity Photos: Real photos used to create fake author profiles for AI-generated articles.

Human Factors

• Confirmation Bias in Audience: Voters readily shared the fake story because it aligned with their beliefs.
• Poor Digital Literacy: Users failed to notice telltale signs of AI manipulation in the video.
• Malicious Intent of Operator: John Dougan weaponized AI to run elaborate Russian disinformation campaigns.

Process and Methods

• Ineffective Platform Moderation: X and Telegram failed to quickly detect and remove viral deepfake content.
• No Human in the Loop: Automated publishing allowed raw AI content to go live with prompt leaks.
• Flawed Search Indexing: Microsoft MSN aggregated and indexed the fake story as a top search result.

Regulatory Environment

• Weak Platform Regulations: Lack of strict regulatory enforcement on social platforms hosting deepfakes.
• Absence of Deepfake Laws: No robust legal frameworks to prevent or prosecute cross-border AI fakes.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide highly detailed, consistent, and cross-verified accounts from multiple credible sources (BBC Verify, NewsGuard, Recorded Future, CNN, CBS). The role of AI (ChatGPT and deepfakes) is explicitly documented, and the perpetrator's identity and methods are well-established.
• Ambiguities identified: The exact deepfake generation software used to create the fake Bugatti employee video is not specified.
• Alternative interpretations: None. The consensus across all reports is that this was a coordinated Russian-linked disinformation campaign.

A coordinated Russian-linked disinformation campaign used ChatGPT and deepfakes to fabricate a story about the Ukrainian President's wife, aiming to erode Western political support. Reaching millions, it highlights the growing threat of AI-scaled foreign influence operations targeting democratic processes and international alliances.

• Overall national security impact: Substantial
• Response level: Substantial
• Scope: Multiple nations
• Primary target: Ukraine
• Other affected: United States, United Kingdom, France
• Alleged perpetrator: Russian-linked network led by John Mark Dougan

Threat characteristics

• Imminence: Long-term. Represents an ongoing strategic information warfare threat and capability that persists across election cycles and geopolitical conflicts.
• Autonomy: Human-controlled. AI systems were strictly utilized as tools by human operators who directed the content generation, narrative formulation, and distribution.
• Novelty: Evolved capability. While disinformation is an established threat, the automated scaling using LLMs combined with deepfake video integration represents a highly evolved capability.

Impact by dimension

• Physical security: Negligible. No physical systems, critical infrastructure, or human safety was threatened or impacted by this digital disinformation campaign.
• Information security: Substantial. Highly coordinated foreign disinformation campaign utilizing AI-generated text and deepfakes to target political leadership and influence Western public opinion regarding military aid.
• Sovereignty: Substantial. Targeted democratic processes in the US and UK ahead of elections by injecting synthetic propaganda designed to influence policy decisions on foreign aid to Ukraine.
• Economic security: Minor. Limited to minor reputational and legal mitigation costs for private entities like Bugatti, with no systemic threat to strategic industries or financial systems.
• Societal stability: Minor. Targeted reputational harm to an individual and localized manipulation of social media users, without causing large-scale civil unrest or systematic rights violations.