In August 2026, a video circulates on social media. It shows a Black man, presented as a migrant recently arrived in England. Facing the camera, he seems to explain that he wants “his money”, “a free house” and an “English wife”.

The video was posted on 4 August 2026 on Instagram and Facebook by a certain “Darren Fitton”, who regularly shares discriminatory content and says exactly what some people already expect to hear. The problem is that it is completely false…

The same profile also posted, on 9 August 2026 on Facebook, an AI-generated image of the hooded man seen in the video. The image came with the same quote about money and free housing as the one in the viral clip.

The face is indeed that of a real migrant. But the original footage was filmed in Tunisia in February 2024. The man, identified as Ahmed, is Sudanese and spoke Arabic. In the original video, he was reacting to a controversial migration deal recently concluded between Albania and Italy. The viral version made him say something entirely different. His voice had been manipulated using artificial intelligence.

This case says something important about our time. Disinformation can now take a real person, a real event and a real voice… and then assemble them to fabricate a false story. The problem is no longer only the fabrication of fakes, but also our relationship to reality, to those who tell it, and to the representations we build from fragments of images.

In many cases, disinformation works precisely because it maintains a relationship with something real. It borrows an existing image, a well-known figure, a crisis that is actually under way or a concern already present in society. It does not create an imaginary world, but rearranges elements of the real world to produce a particular interpretation of it.

In South Africa, amid growing tensions around immigration, Africa Check documented several of these mechanisms in 2026. Old images were recycled to illustrate recent events, while some AI-generated creations claimed to show migrants or politicians in situations that never happened.

Some images are more easily instrumentalised than others.

When a video shows a migrant who has come to claim welfare benefits, the problem lies not only in the authenticity of the video but also in everything this image activates. It meets already established representations of the dependent African foreigner and of the demographic threat he is said to embody. A manipulated image can therefore be considerably effective without being particularly sophisticated: sometimes it only needs to confirm a story its audience already believes.

Societies never look at images with a perfectly neutral eye. We interpret them from our knowledge and our experiences, but also from a set of representations that precede us. Some populations have long been portrayed through recurring narratives. The Black person reduced to poverty or violence. The migrant presented as an anonymous mass. The Muslim associated with threat. The African continent described as a space of permanent crisis. Diasporas, for their part, are often reduced to a few media figures: the immigrant, the undocumented person, the athlete, the artist, the victim or the activist.

In South Africa, content generated or manipulated in the context of anti-migrant mobilisations helped feed discourses already charged with xenophobia.

We must nevertheless be wary of a reading that would make artificial intelligence the cause of all these phenomena. The tool changes, but the social logics that receive it are much older.

A fake video of a migrant saying he came to get a free house would probably not circulate as readily in a society where representations of immigration were different. An AI image showing an African committing a violent act does not just produce false information: it fits into a long history of representations linking certain populations to dangerousness, incivility or threat.

The cruellest part of this drama is that disinformation acts not only on how a group is seen by others, but also on how it sees itself. It can help reinforce an opposition between local and migrant populations, feed ethnic or political tensions, or be used to discredit an opponent, make people believe a statement that was never made, or fabricate the appearance of popular support.

Why is it so dangerous?

Recent work on political deepfakes in Kenya shows, moreover, that how a video is interpreted depends heavily on the social context and the comments that accompany it: the content is not received in isolation, but within an environment where other people are already telling the public what to believe or suspect.

The circulation of such content then poses a problem that goes well beyond the question of recognising an AI-generated image. We need to relearn how to look at the conditions in which information reaches us. Who is posting? At what moment? With what vocabulary? In what political context? What event does this image claim to show? Is there an earlier version? Do other sources tell the same story? Why this particular footage, and why now?

These questions do not mechanically detect every lie. They nevertheless have an essential virtue: they reintroduce distance between the image and the interpretation we make of it.

How do we fight it?

As we see more and more content of uncertain authenticity circulate, another temptation appears: to believe nothing at all. Yet this generalised scepticism would be another form of weakening.

A society in which nothing can be established with enough certainty also becomes a society in which it is harder to hold anyone to account. The answer is therefore neither credulity nor permanent suspicion. It lies in an ability to rank evidence, to identify sources and to accept that a piece of information may sometimes remain uncertain until more elements allow it to be established.

There are already several tools that can detect the use of AI in a video (even if none of these detectors is 100% reliable, they are very good indicators of probability).

Among them:

• invid-project.eu
• Zerogpt
• Copyleaks AI Video Detector and many more…

For media that work on Africa and its diasporas, this requirement is even more important. Because telling these stories also means meeting a certain resistance caused by narratives already entrenched.

Disinformation ultimately reminds us of something older than artificial intelligence: a media image is never only an image. It is selected, framed, accompanied by a text, placed in a context and interpreted by an audience. It can document reality, but it can also help fabricate the way we look at it.

Some lies do not need to be particularly credible: they only need to resemble what we are already inclined to believe. In the case of Africa and its diasporas, this vulnerability is not new. What changes is the speed at which an image can be hijacked, the ease with which a voice can be replaced and the scale at which the same story can be reproduced.

The challenge is then not only whether we can recognise images made by AI. It is whether we can recognise and resist the imaginaries that speak through them. For a technology can fabricate a false voice in a few seconds. It cannot, on its own, explain why we are ready to listen to it. This is perhaps where our responsibility in the face of disinformation truly begins.