Campaign Featuring Purported AI-Generated Audio Attributed to Omar al-Bashir Spreads During Sudanese Conflict

Beginning in late August 2023, a TikTok channel called The Voice of Sudan circulated purported AI-generated audio claiming to be leaked recordings of former Sudanese president Omar al-Bashir. Experts identified several recordings as voice-cloned fabrications, derived from unrelated broadcasts by a political commentator. The campaign emerged during Sudan's civil war and appeared designed to influence political perceptions. TikTok later removed the account for violating synthetic media policies.

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 synthetic media to spread political disinformation and manipulate public opinion at scale during a civil war.

Additional risk subdomains

• 3.2 Pollution of information ecosystem and loss of consensus reality: The widespread use of deepfakes has degraded the local information ecosystem, leading to a 'liar's dividend' where citizens distrust even authentic recordings.

Causal factors

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

The disinformation campaigns were intentionally created and spread by human actors using post-deployment generative AI tools to manipulate public opinion during the conflict.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The report describes 'AI-generated content such as deepfakes' and 'synthetic media', which fall under Risk Level 3 due to transparency obligations to ensure users are informed of the AI's nature.

AI system and alleged parties

• AI system: unspecified voice conversion software
• AI purpose: Voice Generation; Deepfake Video Generation
• Behaviour type: Tool
• Alleged developer: Unknown voice cloning technology developers, Unknown deepfake technology developers
• Alleged deployer: The Voice of Sudan
• Alleged harmed parties: Truth, Omar al-Bashir, National security and intelligence stakeholders, General public of Sudan, General public, Epistemic integrity

Harm severity

Severity ratings are based on this scale.

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 Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Substantial, indirect Substantial
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Substantial, indirect Severe
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: Yes

Directly caused: AI-generated deepfakes, including fake audio of military leaders ordering killings and fake videos of political figures, were directly created and spread on TikTok, X, and Facebook, reaching hundreds of thousands of viewers.

Inferred additional harm: It is likely that many more unmonitored malicious deepfakes and synthetic audio clips were circulated on private messaging apps like WhatsApp, reaching millions of users.

Democracy

Reported: Yes

Directly caused: The deepfake campaigns directly targeted political leaders, military commanders, and public figures to manipulate public opinion and political alignment during a civil war.

Indirectly caused: The proliferation of fake content has weakened social cohesion and undermined the ability of citizens to make informed political decisions.

Inferred additional harm: The ongoing manipulation likely further destabilizes the transition to democratic governance and prolongs the conflict by fueling distrust between factions.

Epistemic

Reported: Yes

Directly caused: AI-generated deepfakes fabricated events, such as a secret coup meeting and military orders to kill civilians, misleading hundreds of thousands of social media users.

Indirectly caused: The 'liar's dividend' has created an environment of deep skepticism, causing citizens and politicians to falsely dismiss authentic recordings as AI-generated.

Inferred additional harm: Widespread erosion of shared reality across Sudan, making it extremely difficult for the population to access reliable information about the war.

People affected

• Occurrences reported: 5
• People reportedly exposed: 500000

Potential causes

Management

• Insufficient Local Moderators: Platforms lack content moderators who understand the local Sudanese context.

Technology

• Advanced Voice Conversion Tech: Voice conversion software allows easy cloning of political figures' voices.
• Imperfect Detection Software: Publicly available detection tools produce frequent false positives.
• Low-Resource Dialect Limitations: AI models struggle with Sudanese dialects, causing robotic tones.

Data Inputs

• Lack of Localized Training Data: Low-resource languages like Sudanese dialects lack diverse training datasets.
• Compression and Low Resolution: File compression and low resolution degrade detection tool accuracy.

Human Factors

• Democratization of Deepfakes: Average users can quickly create and share sophisticated fake audio.
• Unverified Sharing by Leaders: Politicians and public figures share unverified deepfakes on social media.
• The Liar's Dividend Effect: Politicians falsely claim real recordings are AI-generated to escape blame.

Process and Methods

• Absence of On-The-Ground Reporting: Lack of verified local reporting allows online disinformation to spread.
• Inefficient Detection Access: Reliable deepfake detection tools are out of reach for the general public.
• Poor Content Moderation Enforcement: Platforms fail to enforce deepfake policies in regions like Sudan.

Regulatory Environment

• Lack of Platform Accountability: No strong regulatory enforcement to hold platforms liable in conflict zones.

Information quality

• Classification confidence: High
• Reason for confidence: The reports from reputable sources (BBC, African Arguments) provide clear, consistent details about the nature of the deepfake campaigns, the platforms involved (TikTok, X, Facebook), and the specific political figures targeted. The role of AI (voice cloning and deepfake generation) is explicitly stated, and the societal impacts are well-documented.
• Ambiguities identified: The exact software tools used to generate the deepfakes are not identified, and the specific actors behind the 'Voice of Sudan' TikTok account remain anonymous.
• Alternative interpretations: Some might view the satirical deepfakes as harmless entertainment, but the reports show they quickly morphed into political disinformation.

During the Sudanese civil war, various actors deployed AI-generated synthetic media, including voice cloning and deepfakes, to impersonate political and military leaders. This has severely degraded the local information ecosystem, creating a 'liars dividend' where authentic information is dismissed, complicating conflict resolution and destabilizing societal trust.

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

Threat characteristics

• Imminence: Long-term. Represents an ongoing strategic concern regarding the weaponization of synthetic media in conflict zones rather than an immediate, discrete crisis.
• Autonomy: Human-controlled. AI tools (voice cloning, deepfake generation) were used as direct tools by human actors to generate content, with no autonomous decision-making by the AI itself.
• Novelty: Evolved capability. Represents a significant evolution in the scale, accessibility, and sophistication of political disinformation campaigns using generative AI.

Impact by dimension

• Physical security: Negligible. No physical security threats, kinetic attacks, or critical infrastructure compromises were reported in connection with this incident.
• Information security: Substantial. Synthetic media, including voice cloning and deepfakes, was used to impersonate military and political leaders, spreading disinformation that directly impacted the information ecosystem during an active civil war.
• Sovereignty: Substantial. Deepfakes targeted the head of the Sudanese Armed Forces and former leaders, aiming to manipulate political alignment and destabilize government functions during a civil war.
• Economic security: Negligible. No significant economic or technological security impacts were reported.
• Societal stability: Substantial. The proliferation of deepfakes and the resulting 'liars dividend' eroded public trust, undermined social cohesion, and complicated efforts to access reliable information during a humanitarian crisis.