A Crisis in Images: AI-Generated Floods and Journalism’s Watchdog Moment
On the morning of 26 August 2026, the first sign that something apocalyptic was happening in Nepal arrived through a fixed, unblinking camera. CCTV footage from the Rasuwagadhi border point showed a wall of water and mud—moving with the force of a tsunami—sweep over the crossing, carrying rock, timber, and the ordinary structures of everyday life along with it. The clip was raw, grainy, and without commentary, but it did not need any. In the hours that followed, the disaster was given a human face by visual journalists who had made their way toward the wreckage. Associated Press photographer Niranjan Shrestha and Reuters’ Navesh Chitrakar were among the Nepal-based professionals whose images documented not just the immense force of the flood, but also the quieter, more exhausting aftermath: families wading through muddy streets, rescue workers straining against impossible conditions, survivors standing where their homes used to be. Their photographs were a reminder of the power of visual journalism to hold a mirror to the human condition in times of crisis. They were also evidence—real, physical, verifiable evidence of what happened in a particular valley on a particular morning. That mattered more than anyone realised at the time, because within hours, the same social media platforms carrying this essential reporting were also carrying something else. A second wave began to circulate: pictures and videos that looked like Nepal’s flood, yet were not what they seemed. Some were old clips from disasters in other countries, recaptioned and repackaged as breaking news. Others were generated entirely by artificial intelligence, showing scenes so dramatic and slick that they might have been pulled from a disaster film. The water had been joined by a flood of fakes.
The flood of fake content took many forms. Some of it was not invented at all, only recycled. A viral video posted on X appeared to show Nepal’s flood, but was actually footage from a 2025 mudslide in India. Another Facebook post, which Australian Associated Press fact-checkers ran through a Google reverse image search, turned out to contain footage from a flood in Japan in 2021—yet it had already been viewed more than 360,000 times. The same pattern repeated itself across social media: old imagery from Chile, India, and Alaska was being repurposed to illustrate a disaster that had occurred in Nepal. Other fakes were more targeted and more heartbreaking. AI-generated images of missing persons began to appear, bearing the names and faces of real individuals, raising false hopes and potentially diverting rescue efforts. The Singapore Red Cross warned that such images could interfere with emergency operations by causing confusion among families and aid workers. The media ecosystem did its best to keep up. Australian Associated Press, ABC, France 24, and BBC all published fact-checking articles and videos, offering audiences practical tips on how to identify fabrications. France 24 reported on multiple fake videos, including one that had been viewed seven million times on X before it was debunked. The scale of the problem was exhausting even to witness. To scroll through a disaster feed was to be confronted by a chaotic mixture of genuine photojournalism, recycled footage, AI-generated rescue scenes, and captions that could no longer be trusted. In that chaos, the need for verification itself became a form of journalism.
The messy reality of platform enforcement became especially clear in one striking example. A Facebook post on the evening of 27 August showed a heroic-looking rescue scene. AAP flagged the image as AI-generated. The author of the original article also ran it through Sightengine, an AI detection tool, which returned a 99 percent probability that the image was artificial. Yet Facebook did not label it as AI content. The same image was then posted on another Facebook feed, in a different context and with different framing, and it continued to be liked and shared. Meanwhile, two other images—both AI-generated—were correctly labelled by Facebook, revealing just how inconsistent the detection systems remain. Perhaps the most unsettling element was a caption attached to one such image: “this is what resilience looks like.” But if the image is not a photograph, if the scene never happened, then the story being told is not resilience at all—it is simulation. The authorities in Nepal understood the danger. The Press Council launched a new website called “Special Complaints on the Bhotekoshi Incident,” where citizens could report fake imagery and other forms of misinformation. On its first day, more than 100 complaints were filed, and the Council instructed 72 social media users to take down content. The numbers made clear that the problem was not peripheral or accidental. It was structural. And it required an institutional response.
Journalism has always depended on verification. Fact-checking is one of its oldest and most fundamental practices, even if it has only recently moved from the quiet corners of the newsroom into the public eye. In the so-called “post-truth” era, beginning around 2016, fact-checking became a visible and contested front in journalism’s battle against accusations of “fake news” and the broader spread of disinformation. The Nepal flood brought that battle into sharper focus. When news organisations took the time to explain how they knew an image was fake—when they showed the reverse image search, the metadata, the AI detection output—they were doing more than debunking a single post. They were demonstrating that journalism is a discipline, not just a platform. They were also performing a public service: teaching audiences how to be skeptical without slipping into cynicism. This is no small task. In a moment when anyone can create an image and circulate it around the world in seconds, the value proposition of journalism shifts from being first to being certain. The journalist’s role becomes not only to witness but to verify; not only to report but to contextualise. That is why the flood of AI-generated imagery might, paradoxically, reinvigorate the watchdog function of the press. Watchdogs exist to expose abuses of power, but they also exist to protect the ground on which truth sits. In a disaster, the visual record is part of the public commons. If it is polluted by synthetic images, the ability of society to respond with empathy, clarity, and urgency is diminished. By stepping into that gap, journalism can reclaim its purpose.
Why does a fake photograph matter so much? Because visuals are often the first point of entry into a story, and we are, as one observer put it, living in the Age of the Image. We are acculturated to receive our news in pictures, both still and moving. Seeing is believing. It is our first instinct, an ancient reflex developed long before the invention of the camera, much less the algorithm. Yet reading photographs is relatively automatic and cognitively cost-free. We tend to look at an image and accept it without the same careful scrutiny we would give to a written argument. This instinct is deeply human, but it is also exploitable. In the digital media landscape, images are consumed at breakneck speed, and attention spans have been shrinking. The average attention span on digital devices has fallen from about 75 seconds in 2012 to about 47 seconds in 2024—not enough time to inspect a scene for subtle asymmetries, warped hands, unnatural lighting, or the other telltale signs of AI generation. It is even less time to perform a reverse image search or to wonder whether a video might have come from another country and another year. When synthetic images flood a crisis, they consume the same finite attention that ought to be directed at the real event. Issue salience—the sense that a problem is urgent and deserves a response—begins to erode. The real and the fake compete for the same space on the timeline, and after a while, audiences may not know which to mourn, which to share, or which to respond to. Few people verify the things they consume, like, or share; this is not merely a character flaw, but a design feature of platforms that reward speed over accuracy. The result is a public that is simultaneously overwhelmed and underinformed.
The consequences of this misinformation flood are not abstract. Fake content can amplify panic and distress in affected communities, where people are already struggling to make decisions about their safety. It can interfere with rescue services, diverting emergency workers to locations that exist only in pixels and wasting time and fuel in a crisis where every moment matters. It can undermine the charity sector, as donors begin to wonder whether the appeals they see are genuine or manufactured for attention. And it erodes trust in visual journalism, making audiences less certain about the work of professionals who risk their lives to document disasters. None of this is entirely new. Photographs have been manipulated for as long as photography has existed, and the darkroom has given way to Photoshop, and Photoshop has given way to generative AI. But the tools have changed, and the scale has changed. AI images are produced in fractions of a second, often with a level of detail that fools even trained eyes. Combined with shrinking attention spans and an instinctive belief in what we see, these images are becoming increasingly difficult to detect. Yet there is an opportunity hidden in the confusion. The more AI-generated imagery proliferates, the more society needs journalists who are willing to check, verify, explain, and contextualise. The Nepal flood may come to be seen as a moment when journalism’s watchdog role was reinvigorated—not merely as a check on power, but as a guardian of reality itself. In an age of synthetic images, the journalist’s most important credential may no longer be access, speed, or even objectivity. It may be the simple, stubborn insistence that facts matter, and that images, no matter how beautiful or terrible, deserve to be looked at twice.