AI Pop Culture: When Synthetic Content Rewrites History

UNESCO defines “synthetic content” as any audio, image or video created by an AI system, a definition that covers both something invented from scratch and the colorization of a real photo, that is, an alteration of something genuine. It is this second category that complicates verification, because the starting point is not an invention but an actual document.

A case gathered by Lead Stories in August 2026 shows the difference. A black-and-white photo from 1944, held in the archives of the Imperial War Museums, shows children at Fen Ditton Junior School sitting at their desks saying grace before their mid-morning milk. It circulated online in an AI-colorized version, and at first glance it looks like a simple restoration. Once uploaded to Sightengine, though, the image came back as 99% likely to be artificially generated: the cups on the desks, which varied in the original, had all become identical with the rolled edge typical of disposable paper cups; the lighting was uniform in a way that isn’t natural; the faces had been reinvented to the point of resembling other people; the lettering on a classroom poster stayed legible in shape but not in content.

Practical example: before sharing a historical photo billed as “colorized” or “restored” by AI, look up the original black-and-white version in the cited archive, museum, library or press agency, and compare the material details, cups, signage, faces, not just the color.

When the fake isn’t even AI-generated

Not all visual disinformation starts with a generative model. Poynter documented a video of a large explosion shared on X as recent proof of the war in Iran in 2026: it was actually footage of a 2015 incident at a chemical warehouse in Tianjin, China. Likewise, a photo attributed to current fighting was in fact a Reuters shot from May 21, 2023, showing a Hezbollah military exercise.

The distinction matters because the countermeasures differ. Recycled footage or photos get exposed by tracing them back to the source; content that is truly AI-generated leaves other kinds of clues. Google, for instance, embeds a watermark called SynthID in images produced by some of its own tools, which Gemini can detect; videos made with Veo, again according to Poynter, run for about eight seconds at most, a technical limit that makes a longer “generated” clip suspicious. Each of these clues covers only part of the tools available, so it needs to be paired with other checks.

Why history and pop culture make an easy target

Historical images and iconic figures give an edge to whoever generates synthetic content: they are already familiar, so a plausible alteration slips by more easily than an invention made from nothing. A photo of English schoolchildren during the Second World War, or a still from a well-known film, arrives with a built-in frame of trust that people rarely question each time it gets shared.

The Reuters Institute for the Study of Journalism, in its Digital News Report 2026, records a side effect of this pattern: trust in AI chatbot answers about the news sits at 20% globally, against 37% placed in news overall, dropping to 6% in the UK. The same report notes that concern about fake news rose by at least 4 percentage points over the past year in every Western European market, reaching 9 points in Belgium. The same dynamic has a financial side. UNESCO cites a 2024 Statista survey showing that 46% of surveyed fraud experts had encountered synthetic-identity fraud, 37% voice deepfakes and 29% video deepfakes; in a case documented in January 2024, scammers impersonated a company’s CFO on a deepfake video call, convincing an employee to transfer $25 million.

What to do about it

Verification remains more of a craft than an automatic process. Organizations such as PolitiFact, Snopes, BBC Verify, Full Fact, Lead Stories and AFP systematically check viral content against original sources. Anyone relying on a chatbot for a quick check should keep in mind a limit the same experts quoted by Poynter point out: the accuracy of these answers depends on the quality and freshness of the data the model was trained on, as well as on how the question is phrased.

To understand more about the models behind this kind of content, it can help to look at what generative artificial intelligence is and how to spot fake news more broadly. The topic also touches regulation: the European Union addresses the transparency of synthetic content in the AI Act, while broader questions about the limits of today’s models sit at the center of the debate over artificial superintelligence.