The atrocity imagery from el-Fasher is AI-generated, so the reported massacre either did not happen or has been wildly exaggerated

The fakes and the evidence are not the same kind of picture

Line up every item the fact-checkers traced. A mother cowering with three children. A mother and child at the moment of execution. An elderly woman weeping at the camera. They are ground-level, human-scale, close, and legible to anyone in any language within two seconds.

Now line up the material the genocide finding actually rests on. Clusters of pale objects beside patches of soil that shift from red to brown across successive images. A raised sand berm drawn around the perimeter of a city. Earth disturbances at a mosque and a former children's hospital. Fresh burial mounds appearing in civilian cemeteries within two days. It is overhead, geometric, undramatic, and unreadable without a stated method.

These two sets have almost nothing in common, and their distribution is not an accident. Fabrication went where demand was highest and supply was lowest: the world wanted to see the massacre and there was almost nobody inside el-Fasher free to photograph it, so the gap filled with what generative tools are good at, which is exactly the close, emotional, human register. The evidence, meanwhile, sits where fabrication is hardest. Nobody is spoofing a time series of commercial satellite acquisitions, because the imagery does not come from the person making the claim. It comes from a satellite operator, with its own acquisition record and its own dating, and it is bought and cited by name.

That is why the sceptical argument feels strong and is weak. It is an argument about one category of picture being used to dismiss a different one, and the two were never doing the same job.

What is actually on the record

The el-Fasher finding did not appear after the city fell. It was forecast. In July 2023, more than two years earlier, Yale's Humanitarian Research Lab warned the US government, the public and the UN Security Council that a genocidal massacre could occur if the city fell to the RSF. During the eighteen-month siege that followed, the lab published more than sixty-five reports documenting killings and other atrocities against Fur, Zaghawa and other Indigenous non-Arab people.

After the city fell on 26 October 2025, the lab identified at least 150 clusters of objects consistent with human remains in and around el-Fasher between then and 28 November. It set out how it reached that reading: the size of the clusters, the timing of their appearance relative to the capture, and their proximity to reddish ground discoloration that darkened to brown across later images, a progression consistent with blood oxidation. It separately recorded activity consistent with the burning and burial of bodies near the Saudi hospital and in the First Class neighbourhood, both places where it had already documented killings, and new individual burial mounds in civilian cemeteries between 26 and 28 October.

In February 2026 the UN Independent Fact-Finding Mission for Sudan reported to the Human Rights Council under the title Sudan: Hallmarks of Genocide in El-Fasher. Drawing heavily on that satellite analysis together with open-source material, it concluded that the RSF had specific intent, the legal standard known as dolus specialis, and then acted in a pattern consistent with it. The United States had reached its own genocide determination in January 2025.

None of that chain runs through a viral video. It runs through a named lab publishing a stated method over two years, a UN mission reading it, and a satellite operator selling dated imagery to anybody who wants to check.

The makers were not the ones hiding anything

There is a detail in the VRT and RTVE findings that gets lost when the story is told as a deception campaign, and it changes what kind of problem this is.

The man who made the execution image said he made it with AI. He wrote it in his caption, he put an AI icon in his story, and he left his logo on every copy, which is how VRT found him in the first place. The maker of the sobbing-woman video runs an account openly devoted to this kind of production and has posted a walkthrough of his own prompting. Neither was passing the work off as documentary evidence of a massacre. Somebody downstream removed the caption, the icon and the context, and reposted the file.

That matters twice over. It means the standard advice to check the source would have worked, because the source was honest and easy to find. And it means the material now being used to argue that a genocide was invented was not made to argue anything of the kind. It was made for attention, stripped of its labelling by resharers, and then recruited by people making a case its author never made.

The mislabelling did not run in one direction either. The clip of a mother with three children, reshared as RSF fighters firing near her as the city fell, was authentic footage. DW established it had been online since mid-September, more than a month before the fall, and its Arabic desk concluded from the dialogue that the soldiers were Sudanese army rather than RSF, and were asking the woman where she was from rather than threatening her. She says she is Zaghawa and that the RSF had arrested her husband. A claim that she was later killed was added by resharers with nothing behind it.

Why scepticism arrived pointing the wrong way

For three years the useful public instruction about conflict imagery has been: slow down, that picture may not be what it says. It is good advice, it has saved a lot of people from being fooled, and almost nobody has been told where it stops applying.

El-Fasher is where the absence of a stopping rule becomes expensive. A viewer arrives having correctly learned that emotive images from war zones are often fake, meets a set of images that genuinely are, and generalises. The generalisation is not stupid. It is the same reasoning that worked the previous ten times, applied to a case where the evidence lives somewhere the viewer never looks and would not find persuasive if they did.

The vacuum was manufactured, which sharpens all of this. The berm around the city was built to seal it, aid was blocked, and connectivity was intermittent. A siege that prevents documentation produces, at the far end, an argument that the undocumented thing did not happen. Whether or not anyone intended that, it is a reliable second effect, and it means the people best served by the claim are the people who created the conditions for it.