AI Fake Bird Sightings Threaten Citizen Science | KōJō

AI-edited wildlife photos are quietly corrupting the science of where birds live

You spot a reed warbler. You photograph it badly, because the light was wrong and a twig got in the way. You ask an AI to tidy the image up. You post it to iNaturalist. A researcher, somewhere, uses that record to map the species' range. Except the bird in your cleaned-up photo now has feathers that belong to something else entirely, because the algorithm filled in what it could not see with whatever it had. The Guardian reported this week that scientists are now openly worried this is happening at scale, and that the citizen science databases underpinning real ecological research are starting to absorb the noise.

When AI tidies up a wildlife photo, it sometimes adds features from a different species entirely, and that corrupted record can end up in the scientific literature.
From this read

Key facts

Only 1,400 of more than 610 million images on iNaturalist have been flagged for AI use, meaning the true scale of contamination is unknown.
A false red-winged blackbird sighting in central Brazil was traced to a photographer asking an AI platform to make a photo "look better," which introduced features from a different species.
Researchers published a formal warning in the journal Nature, noting hundreds of fake images have already been discovered on popular species-recording databases.
The western reef heron, normally found in Africa and southern Europe, was genuinely spotted in north Wales in June 2026, illustrating how legitimate rare sightings are celebrated and why false ones cause real damage.
If you record wildlife outdoors in the UK, the data you submit to platforms like iNaturalist or Macaulay Library contributes to research on species range shifts. The accuracy of what you post matters. Find out more about supporting your outdoor life at kojo.life.

What the research actually says

Generative AI tools filling in obscured areas of wildlife photographs can introduce species features that do not belong in the original image. A photographer in Brazil captured an epaulet oriole, then asked an AI platform to improve the image; the result introduced red-winged blackbird markings, creating a false record. Most users are not deliberately deceiving researchers—they are enthusiasts using tools they do not fully understand.

The core concern, laid out in a commentary in the journal Nature and covered by The Guardian, is not primarily about deliberate hoaxes. Outright fakes, as Dr Alexander Lees of Manchester Metropolitan University puts it, are usually obvious. Nobody is submitting a toucan sighting from Siberia and getting away with it. The subtler problem is what happens when a well-meaning photographer asks a generative AI tool to improve a genuine image, and the algorithm, working from its training data, fills in obscured areas with plausible-looking features that happen to belong to a different species.

The Brazil case is the clearest example in the source reporting. A photographer photographed what was actually an epaulet oriole, a common species in South America. They asked an AI platform to make the picture look better. The result introduced colouring and markings consistent with a red-winged blackbird, a North American species with no prior record in that part of Brazil. The sighting was recorded. It was wrong. The photographer almost certainly did not intend to deceive anyone.

That distinction matters enormously. Most of the people using AI to edit wildlife photos are enthusiasts who care deeply about the natural world. They are not adversaries of science. They are, in many cases, its most valuable field contributors. But the tool they are using does not know what a red-winged blackbird looks like in the same way a birder does. It knows what one looks like statistically, based on millions of images, and it will produce that statistical average whether or not it belongs in the photograph.

Why citizen science data is genuinely irreplaceable

Researchers rely on platforms like iNaturalist and the Macaulay Library to track how species ranges shift in response to climate change. These large-scale, long-term, geographically distributed datasets cannot be gathered by professional scientists alone. If AI-enhanced images are submitted at volume, the cumulative effect on range maps and phenology studies could be significant, as researchers work from data that appears plausible but is quietly incorrect.

It is worth pausing on why this matters beyond the immediate frustration of a corrupted record. Citizen science platforms like iNaturalist and the Macaulay Library are not hobby archives. Researchers use them to track how species' ranges are shifting in response to a changing climate. Are birds appearing further north than historical records suggest? Are migratory arrival dates moving earlier in the year? Are species that once overlapped now separating, or species that rarely met now sharing habitat?

These are not trivial questions. They are the kind of large-scale, long-term, geographically distributed questions that professional researchers with limited field time and funding cannot answer alone. As Tony Iwane, iNaturalist's director of community support and a co-author on the Nature paper, is quoted as saying: regular people are posting information that a scientist could probably never get at scale. The platform functions, in his framing, almost like a real-time sensor of what is happening on Earth.

That sensor only works if the signal is clean. One corrupted record in a dataset of millions is not a catastrophe. But if AI-enhanced images are being submitted at volume, and if the 1,400 flagged images represent a small fraction of the actual total, the cumulative effect on range maps and phenology studies could be significant. Researchers would be working from data that looks plausible, passes casual inspection, and is quietly wrong.

Photograph: Gabriel McCallin / Unsplash

The specific risk for UK outdoor people

UK birders submit records to global databases used by researchers at the RSPB and academic institutions. Rare sightings carry cultural weight in UK birding communities, creating incentive to produce clean images of remarkable birds. Beyond data corruption, AI editing tools can erode the discipline of accurate observation—the practice of seeing what is actually there rather than what you hoped to see—without users noticing.

In the UK, rare bird sightings carry particular cultural weight. The western reef heron spotted in north Wales in June 2026 made headlines across birding communities precisely because genuine rarities are celebrated events. That culture of excitement around unusual sightings is also, unfortunately, a culture that creates incentive to produce a good photograph of a remarkable bird. If your image of something rare is blurry or partially obscured, the temptation to clean it up is real.

UK-based platforms and the broader global databases they feed into are used by researchers at organisations including the RSPB and academic institutions across the country. When you submit a sighting from a coastal headland in Cornwall, a reservoir in Yorkshire, or a Highland glen, that record has the potential to appear in peer-reviewed research. The provenance of the image matters in a way it did not when everyone was shooting on film and the main editing tool was a darkroom.

There is also a subtler psychological dimension here. Spending time outdoors recording wildlife is, for many people, a practice that connects them to something real and specific. The discipline of accurate observation, of seeing what is actually there rather than what you hoped to see, is part of what makes it valuable. AI editing tools, even used innocently, can erode that discipline without the user noticing. You stop seeing the twig in front of the bird and start seeing the image you wished you had taken.

What the outdoor community can do about it

Do not use AI image editing tools before submitting wildlife records to citizen science platforms. A blurry photograph of a genuine bird is more scientifically valuable than a polished image partly reconstructed by an algorithm. Keep unedited originals for submission; post edited versions to social media separately. Experienced users should flag implausible or suspiciously perfect sightings to maintain data integrity.

The practical ask from researchers is straightforward: do not use AI to edit images before submitting them to citizen science platforms. That includes asking AI tools to remove obstructions, improve lighting, sharpen detail, or otherwise alter the content of a photograph. A blurry record of a real bird is more scientifically useful than a crisp image of a bird that has been partly reconstructed by an algorithm.

If you want to post a beautiful version of an image to social media, that is a separate decision. Keep the unedited original, and submit that. Most platforms accept low-quality images. A record with GPS coordinates, a timestamp, and a genuine photograph, however imperfect, is exactly what researchers need.

It is also worth being alert to what you are seeing in other people's posts. The community self-policing function on platforms like iNaturalist depends on experienced users flagging records that look wrong. If a sighting seems implausible, or if an image looks suspiciously perfect, raising a query is not unfriendly. It is how the system is supposed to work.

In practice: your next outing

Save photographs in the highest-quality uncompressed format available. Treat AI identification apps like Merlin as prompts for your own observation, not confirmed records. Photograph what you actually see—partially obscured birds and distant sightings are valid. Submit original files with contextual notes on habitat, behaviour, and time. If uncertain about identification, state that honestly rather than guess with confidence.

Before you go out, set your camera or phone to save in its highest-quality uncompressed format. Raw files or full-resolution JPEGs give researchers more to work with than compressed social-media exports. If you use a birding app that identifies species from photographs, treat its output as a prompt for your own observation, not a confirmed record. Apps like Merlin are useful tools, but the identification you submit should be yours.

When you are in the field, photograph what you actually see. A partially obscured bird, a bird in flight, a bird at distance, these are all valid records. Note the habitat, the behaviour, the time of day. That contextual information is often as valuable to researchers as the image itself, and it is information no AI can fabricate from a photograph.

When you get home, submit the original file. If the platform allows notes, add them. If you are uncertain about an identification, say so. An honest "probable reed warbler, not certain" is more useful than a confident misidentification. And if you do edit an image for social sharing, be explicit about it. The birding community is generally good at this when the norm is clearly established.

The broader habit here is one that anyone who spends time outdoors recognises: paying attention to what is actually in front of you. That is the practice. The photograph is secondary.

Attention, stress, and the mental cost of being online

AI-generated and AI-enhanced wildlife images degrade the signal-to-noise ratio on birding forums, requiring more cognitive effort to distinguish what is real. This accumulating effort contributes to low-grade background stress that disrupts sleep and raises baseline cortisol. Outdoor wildlife recording serves as a counterweight to this cognitive load, offering restorative attention qualitatively different from scrolling feeds. Remain deliberate about platform engagement whilst keeping actual practice central.

There is something worth naming in the background of this story. The rise of AI-generated content across social media is not only a data quality problem for science. It is a cognitive environment problem for the people who use these platforms. When a significant proportion of wildlife images online are AI-generated or AI-enhanced, the experience of scrolling through birding forums changes. The signal-to-noise ratio drops. Distinguishing what is real requires more effort. That effort accumulates.

People who spend time outdoors recording wildlife often do so partly because it is a counterweight to exactly this kind of cognitive load. The attention required to find and identify a bird in real habitat is qualitatively different from the attention required to scroll a feed. One is restorative. The other, over time, may contribute to the kind of low-grade background stress that disrupts sleep and raises baseline cortisol, something worth reading about if you find yourself more wired in the evenings than you used to be: sleep cortisol and why you're wired at 11pm covers the physiology in detail.

The answer is not to leave the platforms entirely. The data they generate is too valuable. But being deliberate about how you engage with them, and keeping the actual outdoor practice central rather than the social media dimension of it, seems like the right orientation.

Staying sharp outdoors: what daily nutrition has to do with it

Accurate field observation demands sustained attention—scanning habitat, holding potential identifications in working memory, and making confident calls under time pressure all draw on neurological resources affected by sleep, stress, and nutrition. Rhodiola Rosea and DHA support cognitive function during demanding tasks. These are background supports for a life already structured around outdoor practice, not replacements for time in the field.

Accurate field observation requires sustained attention. That is not a metaphor. The cognitive work of scanning habitat, holding a potential identification in working memory while checking field marks, and making a confident call under time pressure draws on the same neurological resources as any other demanding task. Those resources are affected by sleep quality, stress load, and baseline nutrition.

Rōnin includes 350mg of Rhodiola Rosea extract per daily serving. If you want to understand what the evidence actually says about that ingredient before deciding whether it is relevant to you, rhodiola rosea supplement uk stress is a thorough read. The formula also includes 250mg of algal DHA. [GB-NHC] DHA contributes to the maintenance of normal brain function (authorised at ≥250 mg/day; Rōnin delivers exactly 250 mg). And for days when the field session runs long and the mental load is high, best supplements for stress anxiety uk covers the broader evidence landscape honestly. None of this replaces being outside. It is background support for a life that is already built around it. More at kojo.life.

Questions UK outdoor people are asking

The concern applies to both global platforms like iNaturalist and UK-specific schemes; records submitted by UK users feed into the same databases researchers draw on. AI identification tools like Merlin analyse photographs without altering them, so using them is distinct from the editing problem. Basic in-camera processing and standard photo software do not introduce species-feature contamination; the risk lies specifically with generative AI tools asked to improve images.

Does this affect UK-specific platforms, or just global ones like iNaturalist?

The Nature commentary cited in the Guardian piece focuses on iNaturalist and the Macaulay Library, both of which have substantial UK user bases and UK data. Records submitted by UK users feed into the same global databases that researchers draw on. UK-specific recording schemes, such as those run by the British Trust for Ornithology, have their own verification processes, but they are not immune to the same pressures. The concern is platform-wide.

If I use an app like Merlin to identify a bird, is that the same problem?

No, these are distinct issues. AI identification tools like Merlin analyse your photograph and suggest a species. They do not alter the image you submit. The problem described in the Guardian piece is specifically about AI image editing tools that change the visual content of a photograph before it is submitted as a record. Using an AI tool to help identify a bird is fine; using one to make the photograph look better before submitting it is where the risk lies.

What if I genuinely cannot tell whether my photo has been altered by AI?

If you used a standard camera or phone in automatic mode and did not run the image through a generative AI tool, it has not been AI-altered in the relevant sense. Basic in-camera processing, exposure correction, or cropping in standard photo software does not introduce the kind of species-feature contamination described in the source reporting. The concern is specifically about generative AI tools, including chatbots asked to "improve" an image, not conventional photo editing.

My honest take

The trajectory matters more than current scale. With generative AI becoming faster and more embedded in every phone, the ratio of corrupted records will likely worsen. Well-intentioned enthusiasts—serious birders wanting clean images of rare finds—cause unintended harm by not understanding their tools. The fix is simple: keep originals, submit originals, be honest about uncertainty. Accurate observation of the natural world requires presence, patience, and honesty about what you see.

I find this story genuinely unsettling, not because of the scale of the problem as it currently stands, but because of the trajectory. Right now, 1,400 flagged images out of 610 million sounds manageable. In two years, with generative AI tools becoming faster, cheaper, and more embedded in every photo app on every phone, that ratio could look very different.

The people most likely to be affected are the ones who care most. Serious birders who want a clean image of a rare find. Wildlife photographers who have spent hours in a hide for a single shot and want it to look its best. These are not bad actors. They are enthusiasts with a new tool they do not fully understand, which is the most common way that well-intentioned people cause unintended harm.

The fix is not complicated. Keep the original. Submit the original. Be honest about uncertainty. These are habits that good field naturalists already have. The ask is just to extend them to the digital workflow as well.

What I keep coming back to is the deeper value of the practice itself. Accurate observation of the natural world is one of the few activities that genuinely requires you to be present, patient, and honest about what you are seeing. AI editing tools, at their worst, let you substitute the image you wanted for the image you took. That is a small corruption of something important. Worth resisting.