Purported AI-Generated Content Circulates Widely in Sudan Amid Civil Conflict and Information Vacuum

Purported AI-generated deepfakes have circulated widely on Sudanese social media during the ongoing conflict, allegedly spreading false claims, impersonating public figures, and distorting political discourse. Reported incidents include fabricated videos and audio used to mislead or satirize, some of which gained traction among journalists and officials. Observers warn that poor detection tools and limited moderation may be enabling the manipulation of public trust.

Risk classification

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

The incident involves coordinated campaigns using AI-generated deepfakes to spread disinformation and influence public opinion at scale during a national conflict.

Additional risk subdomains

• 3.2 Pollution of information ecosystem and loss of consensus reality: The widespread use of deepfakes and the resulting 'liar's dividend' have severely degraded the information ecosystem in Sudan, undermining shared reality.

Causal factors

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

The deepfakes were intentionally created and spread by human actors to conduct psychological warfare and manipulate public opinion after the AI models were deployed.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The report describes AI-generated content such as deepfakes, which require specific transparency obligations to ensure users are informed.

AI system and alleged parties

• AI system: unspecified
• AI purpose: Deepfake Video Generation; Voice Generation
• Behaviour type: Tool
• Alleged developer: Unknown voice cloning technology developers, Unknown deepfake technology developers
• Alleged deployer: Various actors aligned with Sudanese factions
• Alleged harmed parties: Political leaders of Sudan, National security and intelligence stakeholders, Journalists of Sudan, General public of Sudan, 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 Severe
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Minor, indirect Substantial
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Substantial
• Epistemic: direct Minor, indirect Severe
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: Yes, the report describes generative AI being used to produce and disseminate disinformation and hate speech.

Directly caused: Multiple deepfake videos, audios, and images were created to impersonate leaders and spread fake news.

Indirectly caused: These deepfakes were widely shared on platforms like Facebook, TikTok, and X, reaching hundreds of thousands of viewers.

Inferred additional harm: Continued proliferation of malicious synthetic content across Sudanese social media networks, further polarizing the population.

Democracy

Reported: Yes, the report describes deepfakes being used to manipulate political processes and spread fake news about political and military leaders.

Directly caused: AI-generated recordings fabricated a secret meeting discussing a military coup and fake orders from military heads.

Indirectly caused: These fabrications undermine trust in political leadership and democratic transition efforts in Sudan.

Inferred additional harm: Long-term erosion of democratic institutions and trust in governance as public figures use the 'liar's dividend' to dismiss real evidence.

Psychological

Reported: Yes, the report explicitly mentions the use of deepfakes as part of psychological warfare.

Indirectly caused: The psychological warfare campaigns aimed to manipulate public anxiety and sympathy during an active civil war, affecting hundreds of thousands of social media users.

Inferred additional harm: Widespread anxiety, confusion, and fear among the Sudanese population (estimated at 1,000,000 people) due to fake reports of military orders, coups, and bombings.

Epistemic

Reported: Yes, the report explicitly details the erosion of the information ecosystem and the 'liar's dividend'.

Directly caused: Fake images of bombings, fake videos of diplomats, and fake audio of military leaders were widely circulated as real news.

Indirectly caused: The 'liar's dividend' allows politicians to falsely claim authentic recordings are AI-generated, creating an environment of skepticism.

Inferred additional harm: A complete breakdown of shared reality and consensus truth for millions of Sudanese citizens, making it impossible to distinguish real events from fabrications.

People affected

• Occurrences reported: 7
• People reportedly harmed: 1
• People reportedly exposed: 430000

Potential causes

Management

• Inadequate Content Moderation: Platforms lack moderators who understand the local Sudanese context.

Technology

• Inaccurate Detection Tools: Public detection tools produce frequent false positives and lack clarity.
• Impact of File Compression: Compression and low resolution reduce deepfake detection accuracy.
• Weak Low-Resource Voice Models: Voice cloning models are less effective in low-resource dialects.

Data Inputs

• Lack of Sudanese Dialect Data: AI models lack training data on specific Sudanese dialects.
• Biased Training Datasets: Detection tools depend heavily on specific training data quality.

Human Factors

• Sharing of Unverified Content: Politicians and public shared fake media without verification.
• Exploitation of Liars Dividend: Politicians falsely claim authentic recordings are AI-generated.
• Lack of Technical Expertise: General public struggles to interpret complex detection results.

Process and Methods

• Lack of Reliable Detection Tools: Fact-checkers lack access to easy and reliable verification tools.
• Absence of On-Ground Reporting: Conflict prevents direct reporting, increasing disinformation.
• Poor Platform Policy Enforcement: Social media policies are not effectively enforced in Sudan.

Information quality

• Classification confidence: High
• Reason for confidence: The report provides multiple concrete examples of deepfakes used in Sudan, citing specific platforms (TikTok, Facebook, X), view counts (230k), and dates. It also includes perspectives from fact-checkers (Beam Reports) and AI researchers.
• Ambiguities identified: The exact AI models used to generate the deepfakes are not identified, and the true originators of some campaigns remain anonymous.
• Alternative interpretations: Some content was claimed to be satirical, but later morphed into disinformation, showing ambiguity in intent for some creators.

During the Sudanese civil war, AI-generated deepfakes and synthetic audio have been systematically deployed for psychological warfare and disinformation. This includes fake coup recordings broadcast on National TV and fabricated military orders. The resulting 'liar's dividend' has severely degraded the information ecosystem, posing substantial threats to sovereignty and societal stability.

• Overall national security impact: Substantial
• Scope: Single nation
• Primary target: Sudan
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Near-term. The incident represents an ongoing, active information warfare campaign within a hot civil war, requiring continuous monitoring and response.
• Autonomy: Human-controlled. Human actors actively generated and disseminated the deepfakes using AI tools; the AI systems did not act autonomously.
• Novelty: Evolved capability. Represents an evolved capability where deepfakes and voice cloning are systematically integrated into psychological warfare campaigns during active kinetic conflicts.

Impact by dimension

• Physical security: Negligible. No direct physical security or critical infrastructure compromise reported. AI was used for synthetic media generation rather than kinetic or physical system manipulation.
• Information security: Substantial. Extensive use of AI-generated deepfakes and synthetic audio targeting military and diplomatic leaders, including a fake military order with 230,000 views and a fake coup recording broadcast on National TV.
• Sovereignty: Substantial. Fabricated recordings of military coups and leadership orders directly target state authority and government functions during an active civil war, eroding trust in official communications.
• Economic security: Negligible. No significant economic or technological security impacts reported in the incident details.
• Societal stability: Substantial. Widespread dissemination of synthetic media designed to manipulate public opinion and incite violence, leading to epistemic degradation and the exploitation of the 'liar's dividend' by politicians.