Recordings presented as leaked messages from former Sudanese President Omar al-Bashir drew hundreds of thousands of views on TikTok during Sudan’s civil war. An investigation found that the voice was produced using AI cloning, raising concerns about how convincing audio fakes can circulate in a conflict.
Recordings attributed to a former president
Since late August, a TikTok account had purported to share leaked footage of al-Bashir. He was overthrown by the military in 2019, is accused of war crimes, and has not been seen in public for a year. He is believed to be seriously ill.
Experts who examined the material concluded that the recordings were edited with AI to imitate his voice. At least four clips were traced to excerpts from live broadcasts by a blogger. The investigation found no evidence that the blogger helped create or distribute the campaign.
The clips therefore combined existing speech with a synthetic imitation, making ordinary recordings appear to carry a message from a prominent political figure. The account’s framing as leaked footage could make that claim seem plausible to viewers who encounter the posts without context.
Why the campaign matters
The reason for impersonating al-Bashir is unknown. The BBC speculated that the campaign might make listeners believe he was involved in the ongoing civil war, or bolster a particular political viewpoint. Those possibilities remain speculation; the available findings do not establish who was behind the account or what it intended to achieve.
The case illustrates a broader risk from audio and video manipulation tools that are widely available. Hany Farid, a digital forensics researcher at the University of California, Berkeley, told the BBC he was concerned about the democratisation of access to sophisticated manipulation technology. Non-professionals can now create false material without specialized production skills.
When a voice can be cloned and attached to existing speech, a recording’s apparent familiarity is not proof that the named person said it. In a conflict, a false attribution can shape how people interpret a figure’s role or political position, even when the underlying audio began as someone else’s broadcast.
Deepfakes can weaken trust in real evidence
AI experts warn that fake audio and video can spread disinformation, potentially incite unrest, and influence elections. The concern extends beyond the reach of a single post: repeated exposure to fabricated recordings can make audiences less certain about what they hear.
Mohamed Suliman, a researcher at Northeastern University’s Civic AI Lab, said such recordings could create a climate in which even authentic recordings are doubted. That loss of confidence can make it harder for people to assess genuine evidence, because the possibility of manipulation becomes a reason to dismiss real material too.
The distinction matters in public debate. A false clip can mislead viewers directly, while a steady stream of synthetic media can also erode the value of recordings as evidence. The investigation into the al-Bashir clips addressed the voice’s authenticity, but it did not resolve who organized the campaign or why.
Platform response and the limits of attribution
TikTok suspended the account after the BBC alerted the platform. TikTok said the account violated its policies on distributing false content that may cause significant harm and on synthetic media.
The suspension removed the account, but the case leaves a practical challenge: viewers may encounter or share a convincing clip before its source has been examined. Here, investigators could identify the voice as an AI imitation and trace at least four recordings to a blogger’s broadcasts, yet they found no evidence linking the blogger to the campaign.
That distinction is important when assessing manipulated media. Establishing that a voice is fake does not, by itself, establish who made the fake or what they hoped to accomplish. The evidence in this case supports the finding of AI voice cloning and the account’s suspension; the campaign’s motive remains uncertain.