Purported AI-Generated Audio Disinformation Reportedly Attributed to U.S. Ambassador John Godfrey Circulates in Sudan

In April 2023, a reportedly fake audio message impersonating U.S. Ambassador to Sudan John Godfrey circulated on Sudanese social media. Reportedly produced using AI voice-cloning tools, the message portrayed the ambassador outlining strategies to impose secularism through U.S. support. Researchers attribute the clip to Islamist-linked networks seeking to undermine Western and civilian influence during the ongoing civil conflict. The message was quickly identified as inauthentic.

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 political disinformation and influence public opinion at scale during an active 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 trust in all media, polluting the information ecosystem.

Causal factors

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

The deepfakes and fake audio recordings were intentionally created and spread by human actors to influence public opinion and conduct psychological warfare.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The report describes the creation and dissemination of AI-generated deepfakes, which fall under the transparency obligations of the EU AI Act for synthetic content.

AI system and alleged parties

• AI system: unspecified AI voice-cloning tools
• AI purpose: Deepfake Video Generation; Voice Generation
• Behaviour type: Tool
• Alleged developer: Unknown voice cloning technology developers, Unknown deepfake technology developers
• Alleged deployer: Sudanese Islamist digital propaganda networks
• Alleged harmed parties: United States-Sudan relations, Truth, National security and intelligence stakeholders, John Godfrey, General public of Sudan, General public, Epistemic integrity

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. 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 Substantial, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Substantial, indirect Minor
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Substantial, indirect Substantial
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

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

Directly caused: AI-generated deepfakes, including fake videos of the U.S. ambassador and fake audio of military leaders ordering killings, were spread on Facebook, TikTok, and X.

Inferred additional harm: It is likely that numerous other unverified deepfakes and malicious propaganda pieces were generated and circulated across local messaging apps like WhatsApp.

Democracy

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

Directly caused: Deepfakes were used to manipulate public opinion, undermine civilian rule, and support military factions during a transition crisis.

Indirectly caused: The liar's dividend allowed politicians to dismiss authentic recordings of their conversations, undermining political accountability.

Inferred additional harm: The systematic use of disinformation likely delays peaceful democratic transition and deepens polarization among the Sudanese public.

Epistemic

Reported: The report explicitly describes severe epistemic harm and the erosion of shared truth caused directly and indirectly by the incident.

Directly caused: AI-generated fake recordings and images misled political leaders, journalists, and the national TV, who shared them as authentic.

Indirectly caused: The prevalence of deepfakes created an environment of skepticism, pushing citizens to question even valid, authentic information.

Inferred additional harm: The total collapse of trust in digital media and news sources within Sudan, making it nearly impossible for citizens to access reliable information about the conflict.

People affected

• Occurrences reported: 7
• People reportedly harmed: 10
• People reportedly exposed: 500000

Potential causes

Management

• Poor Policy Enforcement: Social media platforms fail to enforce deepfake policies in Sudan.
• Lack of Local Moderators: Platforms lack content moderators who understand the local Sudanese context.

Technology

• Inaccurate Detection Tools: Public detection tools often produce false positives and lack transparency.
• File Compression Vulnerability: File compression and low resolution heavily degrade detection accuracy.
• Imperfect Low-Resource Models: Voice cloning models are less effective on low-resource Sudanese dialects.

Data Inputs

• Lack of Sudanese Dialect Data: AI models lack training data for Sudanese dialects, leading to poor quality.
• Use of Public Voice Samples: Creators use publicly available voice samples to clone target voices.
• Re-used Deepfake Templates: Deepfakes are often re-produced from templates, easing generation.

Human Factors

• Public Susceptibility to Fakes: Politicians and public figures easily fall prey and share deepfakes.
• The Liars Dividend: Politicians falsely claim real recordings are AI to deny accountability.
• Sarcastic Content Creation: Creators make sarcastic fakes that the public mistakes for real news.

Process and Methods

• Absence of On-the-Ground Reporting: Lack of local reporting allows online disinformation to spread unchecked.
• Lack of Reliable Detection Access: The general public lacks access to advanced deepfake detection tools.
• Unchecked Sharing by Media: National TV and journalists share unverified AI recordings.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide clear, detailed accounts of multiple specific deepfake incidents in Sudan, their social and political impacts, and the perspectives of local fact-checkers and researchers. The role of AI as a tool for generating synthetic media is explicitly documented.
• Ambiguities identified: The exact technical platforms or models used to generate the deepfakes are not specified.

The systematic deployment of AI-generated deepfakes by warring factions in Sudan represents a substantial national security threat. By targeting military and foreign leaders, these campaigns conduct effective psychological warfare, disrupt governance, and pollute the information ecosystem. This creates a 'liar's dividend' that severely undermines political accountability and societal stability.

• Overall national security impact: Substantial
• Scope: Multiple nations
• Primary target: Sudan
• Other affected: United States
• Alleged perpetrator: Sudanese political and military factions

Threat characteristics

• Imminence: Long-term. Represents an ongoing strategic concern and persistent information warfare capability rather than an immediate 72-hour crisis.
• Autonomy: Human-controlled. AI tools were directly controlled and utilized by human actors to generate specific synthetic media, with no autonomous decision-making by the systems.
• Novelty: Evolved capability. Demonstrates a significant evolution in the deployment of generative AI for coordinated disinformation and psychological operations in an active war zone.

Impact by dimension

• Physical security: Minor. Fake audio of military head ordering civilian killings and fake images of university in flames were used to incite physical violence, though no direct kinetic AI control occurred.
• Information security: Substantial. Widespread use of sophisticated deepfakes targeting the US Ambassador and Sudanese leaders to conduct psychological warfare and pollute the information ecosystem during an active conflict.
• Sovereignty: Substantial. Deepfakes targeting military heads and political leaders undermine state authority, disrupt civilian transition, and allow politicians to evade accountability via the 'liar's dividend'.
• Economic security: Negligible. No significant threats to economic systems, strategic technology, or financial stability were reported in connection with this incident.
• Societal stability: Substantial. Mass-scale disinformation and synthetic media erode public trust in shared truth, deepening polarization and threatening societal stability during a national crisis.