Generative AI is contaminating citizen science records with fake wildlife images

Generative AI Corrupts Wildlife Records on Citizen Science Platforms

A seemingly remarkable discovery in central Brazil quickly unraveled into a cautionary tale about how generative AI is contaminating scientific data. What appeared to be a red-winged blackbird—a North American species never recorded in that region—was actually an epaulet oriole, a common local bird 1

1

. The photographer had simply asked an AI tool to make the image "look better," but the software rebuilt the bird, adding features from a different species entirely.

This incident sits at the heart of a warning published in Nature Ecology & Evolution by researchers from Cornell and Manchester Metropolitan University 1

1

3

3

. They argue that AI-generated content is starting to pollute citizen science platforms that scientists rely on to understand where species live and how they respond to environmental changes. The problem manifests in two ways: outright fakes created from scratch, and more commonly, well-intentioned edits that inadvertently alter the field marks essential for species identification.

How AI-Altered Images Threaten Biodiversity Research

Platforms like iNaturalist and Macaulay Library are far more than hobbyist galleries. iNaturalist alone holds more than 610 million wildlife images that scientists mine to track how species respond to climate change and habitat shifts 1

1

3

3

. Tony Iwane, iNaturalist's director of community support and co-author of the Nature paper, described the network as "almost like a sensor of what is happening on Earth in real time" 3

3

. But that sensor only works if the information feeding it remains accurate.

When AI-enhanced bird photos enter these databases, they distort real-world data in ways that cascade through ecological research. A single fabricated sighting can suggest a species has shifted its range when it hasn't, leading to flawed conservation decisions 1

1

. Dr. Alexander Lees, the Manchester Metropolitan University ecologist who led the research, told The Guardian that "wildlife photographers can be quite obsessed with getting a beautiful photo, but there's a risk that the image might actually cause problems down the line when AI has been used to edit it" 3

3

4

4

.

Researchers have already identified several hundred suspect images across major databases, though the true scale remains unknown since many slip past unnoticed 1

1

3

3

. On iNaturalist, only 1,400 of its 610 million images carry an AI flag, with roughly 600 marked as fully generated—a figure that's either reassuring or alarming depending on how much remains undetected 1

1

3

3

.

The Danger of Model Collapse in Species Identification

The threat extends beyond contaminating scientific data—it creates a feedback loop that degrades the AI tools birdwatchers depend on. Apps like Merlin use machine learning for species identification, relying on recordings, geographical distribution data, and training images from these same platforms 2

2

. When fake wildlife images corrupt the training data, the algorithms produce unreliable results, making it more likely for even well-intentioned enthusiasts to make false observations. This self-perpetuating problem, known as model collapse or "Habsburg AI," means the more AI-generated content pervades the internet, the more it manifests in degraded quality of future models 2

2

.

Lees described his experience bluntly: "a huge volume of wildlife photos" on Facebook are now simply AI-generated imagery 1

1

3

3

. While outright hoaxes like a toucan sighting in Siberia are easy to spot, the subtle edits pose the real danger 1

1

3

3

. When a birder asks AI to remove a branch or sharpen a blurry shot, the model rebuilds the bird, potentially erasing the precise markings that distinguish one species from another.

Platforms Respond as Conservation Efforts Face New Threats

Citizen science organizations are starting to address the issue, though detection remains challenging. iNaturalist now allows users to flag images in two ways: a fully-AI-generated flag that hides the image entirely, and an over-manipulated flag that drops it to "casual" grade so it never reaches research databases 1

1

. The researchers are calling for stronger image-authentication checks, metadata verification, and above all, education so contributors understand that "improving" a photo can break the science 1

1

.

Separate research published in Conservation Biology warns that fabricated AI wildlife videos are also distorting conservation efforts and public understanding of animal behavior 5

5

. Viral videos showing birds sheltering chicks from rain or predators behaving affectionately with prey misrepresent established facts about species behavior, potentially skewing conservation funding toward "cuter" mammals while less popular groups lose support 5

5

.

The broader implications are stark. As birdwatching forums and other platforms struggle with misinformation, the lesson mirrors challenges across every corner of the internet that AI has touched. Once fakes become indistinguishable from reality, trust becomes the scarcest resource, and the data integrity that underpins ecological research hangs in the balance. Researchers stress that most cases aren't malicious—contributors simply want better-looking pictures—but the consequences for biodiversity monitoring and media literacy are profound 3

3

5

5

.