The War in Sudan: Deepfakes Emerge as a Dangerous Weapon

The use of deepfake technology in Sudan's ongoing conflict has turned social media into a new battleground, where disinformation spreads rapidly, contributing to confusion and distrust in an already fragile political climate. While still in its rudimentary stages, deepfake models are becoming a tool of choice for various actors seeking to manipulate public opinion and influence the war between the Sudanese army and the paramilitary Rapid Support Forces (RSF).

In April 2024, an image of a burning building falsely identified as part of Al-Jazeera University in Wad Madani circulated widely on Facebook, claiming that it had been bombed by the Sudanese army. This incident, though fabricated, was shared by many political leaders and public figures before being debunked. Such cases are part of a broader trend, where AI-generated images and videos are used to sway public sentiment and garner sympathies for one side of the conflict or the other.

The trend toward AI-driven disinformation began early in the current war. In August 2023, the *Daily Mail* uncovered a fake video of the U.S. ambassador to Sudan, claiming that the U.S. had plans to reduce the influence of Islam in the country. In October of the same year, the BBC exposed an AI-generated campaign impersonating Sudan's former leader, Omar al-Bashir, which gained hundreds of thousands of views on TikTok. The rapid spread of such misinformation highlights the impact of deepfakes on the Sudanese information landscape.

By March 2024, deepfake technology had become more sophisticated. For instance, a false recording surfaced on X (formerly Twitter), allegedly featuring the head of the Sudanese Armed Forces ordering the killing of civilians and deployment of snipers. Despite its lack of authenticity, the recording was widely circulated, even being shared by notable Sudanese politicians. This type of deepfake content, designed to incite violence or manipulate political perceptions, exemplifies the dangerous role that AI-generated media now plays in Sudan's ongoing conflict.

In addition to politically motivated deepfakes, there have been attempts at satire that blur the line between humor and disinformation. In September 2023, a satirical deepfake video showed RSF leader Mohamed Hamdan Dagalo (Hemedti) singing a song in support of the Sudanese Armed Forces, a clip that went viral. While viewers recognized the humorous intent, such content can easily be repurposed or misinterpreted, further contributing to the disinformation ecosystem.

Efforts to combat the spread of AI-generated disinformation in Sudan have been limited but are gaining momentum. Beam Reports, a Khartoum-based fact-checking organization verified by the International Fact-Checking Network, has been actively tracking deepfakes in Sudan since 2023. The organization has highlighted the challenges posed by AI, noting that while deepfake technology is used to spread misinformation, there has not been a significant increase in its use over the past six months. However, Beam Reports has seen a rise in misleading audio content attributed to Sudanese public figures.

In a UNESCO-organized online seminar in May 2024, Beam Reports emphasized that the lack of on-the-ground reporting is exacerbating the spread of misinformation, further complicated by the increasing use of generative AI to produce and disseminate false content. Without proper journalism and reliable information sources, disinformation flourishes, amplified by AI technologies that make fake content increasingly convincing.

Several individual fact-checkers have also joined the effort to debunk deepfake content. Mohanad Elbalal, a UK-based Sudanese activist, explained that he uses tools like reverse image searches to detect deepfakes, especially when suspicious content goes viral. However, deepfake detection remains a challenge due to the limited availability of reliable detection tools, as noted by Shirin Anlen, a media technologist at Witness. She pointed out that publicly available detection tools often produce false positives, and the results can be difficult to interpret.

The limitations of current deepfake detection methods are further exacerbated by the low-resource nature of Sudanese dialects. According to Mohamed Sabry, a Sudanese AI researcher, voice-cloning models for Sudanese Arabic are still relatively underdeveloped, making the current generation of deepfakes less convincing. However, as AI technology improves and more resources are dedicated to training models on Sudanese dialects, the quality of deepfakes is expected to improve, raising concerns about the future impact of this technology on Sudan’s information ecosystem.

Beyond the technological challenges, there is also the risk of what is known as the "liar's dividend"—the tendency for real information to be dismissed as fake due to the prevalence of deepfakes. In June 2023, Sudanese politician Mubarak Ardol claimed that a leaked recording of a conversation mocking the army leader was created using AI software. This phenomenon fosters a climate of skepticism, where people begin to question the validity of even legitimate information, further eroding trust in public discourse.

Social media platforms have started taking steps to curb the spread of deepfakes. YouTube, for instance, has a policy prohibiting "technically manipulated content that misleads users and may pose a serious risk of egregious harm." The platform also requires creators to disclose when content has been altered using AI, applying transparency labels to such content. However, these measures are not always enforced effectively in regions like Sudan, where there may be a lack of content moderators who understand the local context.

The growing use of deepfakes in Sudan, coupled with limited detection tools and a strained information ecosystem, represents a significant challenge for the country as it navigates an already complex political conflict. As AI technology continues to evolve, the stakes for managing misinformation and protecting the integrity of public discourse will only increase.