AI Deepfakes Are Destabilizing Africa: From Burkina Faso's Coup to Sudan's War

Imagine watching a video of a military commander ordering the killing of civilians — deploying snipers, occupying buildings, targeting a population. The video is shared 230,000 times. Politicians amplify it. It reshapes how a nation understands its own war.

Now imagine that the commander never said those words. The recording was generated by artificial intelligence. And by the time anyone proves it's fake, the damage is done.

This happened in Sudan. In March 2024, an audio deepfake attributed to the head of the Sudanese Armed Forces went viral on X, appearing to order atrocities. It was viewed 230,000 times and shared by hundreds, including known Sudanese politicians — people whose positions lend credibility to whatever they share. The clip was fabricated. But in a country already torn apart by civil war, truth arrived too late.

Welcome to Africa's deepfake crisis.

Burkina Faso: The AI Coup Machine

One of the first AI video deepfakes to emerge on the African continent followed the second military coup in Burkina Faso in September 2022 — when Captain Ibrahim Traoré seized power.

Shortly after, a series of AI-generated videos began circulating online, first on Facebook, then on WhatsApp groups across the Sahel, and eventually on X. The videos featured individuals who appeared to be Pan Africanists and Americans from Africa, urging citizens to support the military junta.

They were not real people. They were AI-generated avatars.

Investigation by the social media analytics firm Graphika traced the videos to Synthesia, a commercial AI platform that creates realistic talking-head videos using AI-generated personae. The pronunciation errors, visual distortions, and the tell-tale appearance of the avatars matched Synthesia's product line exactly.

The implications were chilling. A commercial AI tool — designed for corporate marketing and training videos — was being used to generate political propaganda supporting a military coup. And the same investigation linked the use of Synthesia to a Chinese influence campaign called "Spamouflage," which uses hundreds of fake accounts to "praise China, criticise the United States, and attack the Hong Kong pro-democracy movement."

The Burkina Faso deepfakes were never definitively attributed to Russia or any specific state actor. But researchers noted that the campaign "bore all the hallmarks of a Russian influence operation" — and Wagner's Africa Corps was operating openly alongside the military junta at the time.

Sudan: When Deepfakes Weaponize War

Sudan's civil war, which erupted in April 2023, has become a laboratory for AI-powered manipulation.

Beyond the fabricated military orders, AI voice-cloning technology has been used to clone the voice of former Sudanese leader Omar al-Bashir — a man charged with war crimes who has not been seen in public for years. The recordings, posted on an anonymous TikTok account, were presented as "leaked recordings" of the ailing leader criticizing the current military leadership.

A BBC investigation found that some of the material originated from old Bashir recordings, while newer segments appeared to be AI-generated. The recordings matched a Facebook Live broadcast by a popular Sudanese political commentator known as Al Insirafi, who is believed to live in the United States. Analysts concluded the motivation was either to legitimize Bashir's political influence from exile or to use his voice to support a particular political faction.

In a country where millions are displaced and information is scarce, these deepfakes do not need to be perfect. They only need to be convincing enough to spread — and in the chaotic information environment of a civil war, the bar for "convincing enough" is terrifyingly low.

The Liar's Dividend

Perhaps the most insidious consequence of Africa's deepfake crisis is what researchers call "the liar's dividend."

In Mauritius, ahead of the November 2024 elections, leaked material emerged — phone calls involving senior politicians, police officers, journalists, and foreign diplomats. The material included what Human Rights Watch described as a leaked conversation in which "a senior police officer allegedly asked a doctor to alter a report about the death of a person in police custody."

The government's response? Dismiss it all as AI-generated fakes.

The Mauritian authorities shut down all social media platforms in the country — a dramatic move that silenced legitimate public discourse along with any potential disinformation. The prime minister's use of the "deepfake defense" backfired spectacularly: the attempt to dismiss real evidence as fake contributed to the PM losing an election that had been expected to be a sure win.

This is the liar's dividend in action. Once deepfakes exist, any inconvenient truth can be dismissed as fabricated. Any recording can be called AI-generated. Any evidence can be denied.

As Terence Corrigan from the South African Institute of International Affairs warns, fragile states in Africa are "especially vulnerable" because "malleable communications and a toxic information environment are frightening resources for opportunistic politicians, identitarian hustlers, and malign external actors."

Nigeria: Deepfakes in Democracy

Nigeria's 2023 elections saw extensive use of AI-generated content. Artificially generated images and videos falsely linked candidates to terrorist groups. Deepfake endorsement videos depicted celebrities like Elon Musk and Donald Trump backing specific candidates. Videos were fabricated to show candidates planning to rig elections.

A TikTok video of Borno State Governor allegedly buying voter cards from displaced persons generated 300,000 reactions and over 60,000 views. The video was a "cheapfake" — a low-quality manipulation — but its effect was multi-layered, undermining both the candidate and the electoral commission simultaneously.

The Arms Race

The technology is accelerating faster than the defenses. The KAS study documents a sobering reality: there is currently no 100% accurate detection method for AI-generated content. Humans perform worse than software at identifying deepfakes, but AI detection software itself produces both false positives and false negatives. The "GAN conundrum" — where AI generators and AI detectors are locked in an evolutionary arms race — means that perfect detection may never be achievable.

Meanwhile, the cost of creating deepfakes continues to fall. Commercial platforms like Synthesia make video deepfakes accessible to anyone with a credit card. Open-source AI models make audio cloning possible with just a few seconds of sample audio. And large language models generate convincing text at volumes that would have required entire newsrooms a decade ago.

For Africa — where 84% of people rely on social media for news, where independent media is under pressure, and where democratic institutions are still consolidating — this is not a future threat. It is a present emergency.

---

This investigation draws on research by Graphika, Clemson University Media Forensics Hub, BBC investigations, Human Rights Watch, Africa Check, Africa in Fact, the Africa Centre for Strategic Studies, and the Konrad Adenauer Stiftung's 2025 study on AI-Generated Disinformation in Europe and Africa.

Revvmark is an independent investigation platform. Read more at revvmark.com

Based on reporting by REVVMARK