For roughly 150 years, a photograph functioned as a kind of evidence — a certificate that something had actually happened. Generative AI didn't invent the fake photo, but it did something older techniques couldn't: it made fabricating a historically convincing image as fast and cheap as typing a sentence.
Confirmed
Manipulated photography is genuinely old. A Rijksmuseum exhibit documents doctored images circulating more than a century ago — a man with a wheelbarrow carrying a fake severed head, postcards of impossibly giant geese — proving the impulse to fake a photo predates AI by generations. What's changed is scale and speed, not the underlying temptation.
In spring 2025, a TikTok trend of AI-generated "history POV" videos — animated first-person scenes set during real historical events — went viral, with individual videos accumulating millions of likes. Academic researchers studying the trend found it spanned from earnest educational reconstructions to content that trivializes atrocity for engagement. In January 2026, a coalition of German memorial sites, including sites tied to former concentration camps, published an open letter demanding that social media platforms take proactive action against what they explicitly termed "AI slop" — mass-produced AI content distorting Holocaust history by sensationalizing it.
This isn't confined to distant history. In early January 2026, AI-generated images falsely depicting the capture of Venezuelan president Nicolás Maduro spread widely across social media, blending fabricated visuals with real footage. Fact-checking organizations debunked the images, but researchers found much of the content kept circulating anyway. A peer-reviewed study published at the CHI 2026 conference tested whether labeling AI-generated images as such actually protects viewers from misinformation — and found a genuine side effect: labels did reduce belief in false claims illustrated with AI images, but participants became more susceptible to false claims paired with human-made images, and more hesitant to believe true claims when the accompanying image happened to be labeled AI-generated.
Genuinely uncertain
Whether the fix is more labeling, better labeling, or something else entirely is not settled. The CHI 2026 finding is uncomfortable precisely because it suggests the most obvious policy response — mandatory AI-content labels, now required under frameworks like the EU AI Act and South Korea's AI Basic Act — can create a false sense of security around unlabeled content rather than simply protecting people from labeled content, without anyone intending that outcome.
It's also unresolved how much this actually changes public trust at a societal level versus at an individual moment of confusion. Separate research from the Reuters Institute and University of Michigan found that exposure to hyperrealistic misinformation breeds what some scholars call "truth fatigue" — a general erosion of confidence in distinguishing fact from fiction — but measuring that erosion precisely, and distinguishing it from ordinary media skepticism that predates AI, remains genuinely difficult.
The Door This Opens
AURORA might tell you this isn't really a story about lies — it's a story about what a society does once it can no longer assume a photograph or video is a trustworthy witness by default, the way it could for roughly a century and a half. ECHO would tell you the deeper danger sits specifically with historical memory: living memory of events like the Holocaust is finite, and once the last witnesses are gone, whatever record remains — AI-distorted or not — becomes the only version future generations inherit. The choice this door leaves isn't whether to use AI to visualize history. It's whether the people building and sharing that content treat the distinction between illustration and evidence as something worth preserving.