Fake Historical Videos by AI Increasingly Convincing
Experts warn that artificial intelligence is not filling gaps in history, but creating content that matches our expectations and stereotypes.
Experts warn that artificial intelligence is not filling gaps in history, but creating content that matches our expectations and stereotypes.
Artificial intelligence is increasingly being used to create fake historical videos depicting events from the eruption of Vesuvius to the Chernobyl nuclear disaster. While some of these creations can appear incredibly convincing, media researchers and historians warn that they do not represent a window into the past, but rather a mirror of our present.
According to a report by Brazilian outlet G1 Globo, media researcher Roland Meyer believes that the trend of creating fake historical footage responds to a specific need. "It seems possible, through AI, to recombine images from the past to generate synthetic images that do not exist, but which might have existed, and thus fill alleged or real 'gaps' in archives and historical records. This seems to me like a deceptive promise," Meyer said.
Historian Jürgen Zimmerer adds that every interpretation of a particular scene eliminates other possible interpretations and depictions that also exist. This artificially simplifies history and strips it of its complexity.
One example is a TikTok video showing Pompeii before, during, and after the eruption of Vesuvius in 79 AD. Although labeled as AI-generated content on YouTube, the video contains typical characteristics of artificial intelligence: people move like zombies, and the appearance of the protagonist does not match the typical visual identity of women in ancient Rome.
Another example includes a video allegedly showing surveillance footage from Chernobyl in 1986. Although it lacks a watermark, users on platform X identified it as AI-generated content through community notes. Similarly, a YouTube video created with the Veo tool shows the raising of the American flag on the Japanese island of Iwo Jima. The original photograph was taken by Joe Rosenthal on February 23, 1945, and the AI version adds details and camera angles not recorded in reality. G1 Globo reports that the YouTube channel that posted the video promotes "patriotic stories" and "content that brings American history to life."
Meyer particularly warns about the problematic nature of such depictions. "AI-generated images of the past primarily reproduce the patterns, clichés, and stereotypes through which that past is usually depicted. They are, above all, images of our expectations: not a window into the past, but a mirror of our present," he told DW, as reported by G1 Globo.
Artificial intelligence models use both authentic sources and created images for their training, including historical photographs, digitized images, scenes from films, and video games. The problem is that AI lacks an understanding of historical context. "It produces something that seems convincing at first glance, not because it is historically accurate, but because it matches our expectations," Meyer explains. There are also warnings that a large volume of AI-generated forgeries could over time replace authentic documents, especially for events like World War II.
Although watermarks and labels on social media are useful, they are often removed, and AI tools are constantly advancing. G1 Globo therefore advises critical analysis: examine where the image comes from, who is presenting it, who it is intended for, and with what purpose. Videos depicting periods before the invention of cameras are obviously fake, but those related to more recent history can be more dangerous because they risk replacing authentic documents.
There are also positive examples of using artificial intelligence. In Pompeii, Italy, archaeologists used AI for the first time to reconstruct the appearance of a victim of the 79 AD eruption. This reconstruction is based on archaeological findings and is presented as a scientific approximation, not an authentic depiction of the person, with the aim of bringing research closer to a lay audience.
Interestingly, the desire for instant knowledge is not new. As Hacker News writes, an essay by Anne Lawrence-Mathers published on July 29, 2026, on Public Domain Review explores the medieval manuscript Ars notoria, which promised rapid mastery of all university subjects through complex diagrams, prayers, and mysterious rituals. The work, whose authorship is unknown, appeared in the 13th century, and 56 manuscripts have survived to this day.
Unlike other magical texts, Ars notoria presented itself as highly devout. Its diagrams did not illustrate the text but functioned almost like religious icons-it was believed that possessing and using them enabled direct contact with benevolent supernatural forces, even God. Rituals included prayers and contact with angels, but church authorities were not impressed. Thomas Aquinas specifically condemned the work in his Summa theologiae, devoting question 96 to it, calling it "unlawful and futile."
