In an age when the boundaries between the real and the generated grow harder to discern, citizen science platforms that have become foundational to global biodiversity research now face a quiet infiltration: AI-fabricated species, indistinguishable to the untrained eye, slipping into databases that inform conservation, policy, and scientific literature. The threat is not entirely new — hoaxers have painted butterflies and invented creatures since the age of Linnaeus — but generative AI has transformed a rare mischief into a scalable crisis. What stands between the integrity of the natural reco
AI-Generated Fake Species Infiltrate Citizen Science Platforms
This mantis does not exist. It's a fabrication, created with generative AI.
Mark
Why does it matter if a fake mantis or bird shows up on iNaturalist? Aren't there enough real observations that a few fakes get lost in the noise?
Mimi
Because scientists build conservation decisions on this data. If you're trying to understand whether a species is expanding its range or declining, and half your data points are fabricated, your conclusions are wrong. And you might not know they're wrong.
Mark
But surely experts can spot a fake. Kris Anderson caught the mantis immediately.
Mimi
He did, because he's spent decades with that genus. But not every species has a Kris Anderson watching. And as the AI gets better, even experts will struggle. The willow tit in Scotland almost fooled ornithologists.
Mark
So what's the solution? Ban AI?
Mimi
No. The platforms are adding flagging tools, and they're working on authentication protocols. But the real answer is human attention—thousands of reviewers catching problems before they spread. The irony is that citizen science only works because citizens care.
Mark
What about using AI to catch AI fakes?
Mimi
That's part of it. But there's a limit. An AI trained to spot fakes can be fooled by a better AI trained to fool it. You end up in an arms race. The human eye, trained by experience, is still the most reliable detector we have.
Mark
Is this going to contaminate scientific papers?
Mimi
Probably, eventually. Lees thinks it's inevitable. If someone submits a fake observation to a journal and it passes peer review, the paper gets published. The fake becomes part of the permanent record. That's the real danger.
El Pulso
- Expert identifiers on platforms like iNaturalist are encountering AI-generated creatures — mantises with impossible anatomy, birds with deformed feet — that have never existed in nature, and the fakes are growing harder to distinguish from the real.
- A single AI-enhanced photograph nearly inserted the first-ever Brazilian sighting of a red-winged blackbird into the scientific record, illustrating how a subtle algorithmic touch-up can corrupt downstream research conclusions.
- Because biodiversity databases now underpin conservation assessments, environmental law, and species surveys, contaminated data doesn't just mislead — it can misdirect resources and policy at a moment when ecological stakes are at their highest.
- Platforms are racing to respond with community flagging tools, image-authentication protocols, and metadata verification, while researchers publish urgent warnings in journals like Nature Ecology & Evolution.
- The paradox at the center of the crisis is that AI, when anchored to human verification, is also one of conservation science's most promising tools — the enemy and the ally are the same technology, separated only by oversight.
In an age when the boundaries between the real and the generated grow harder to discern, citizen science platforms that have become foundational to global biodiversity research now face a quiet infiltration: AI-fabricated species, indistinguishable to the untrained eye, slipping into databases that inform conservation, policy, and scientific literature. The threat is not entirely new — hoaxers have painted butterflies and invented creatures since the age of Linnaeus — but generative AI has transformed a rare mischief into a scalable crisis. What stands between the integrity of the natural record and its corruption is, as it has always been, the patient attention of people who care deeply about the truth of the living world.
When Kris Anderson, a mantis specialist with more than 26,000 identifications on iNaturalist, encountered a user-submitted image of a supposed Malaysian Deroplatys, the head shape was wrong, the antennae matched nothing in his catalog, and the body patterns belonged to no known species. The mantis had never existed. It was a product of generative AI — born from code, not evolution.
Anderson is part of a growing community of experts watching fake species accumulate across citizen science platforms. In November 2025, a photograph of a willow tit appeared on a Scottish birding site with deformed feet and beak — entirely fabricated. Alexander Lees, a biodiversity researcher at Manchester Metropolitan University, has been documenting the pattern. In July, he and colleagues published a warning in Nature Ecology & Evolution: citizen science databases risk becoming contaminated. To illustrate the ease of the problem, they asked an AI to enhance a real photograph of a South American oriole. The system added a red flare to the bird's wing — subtle enough to fool an expert ornithologist or a computer-vision classifier into registering a different species. That precise scenario had already played out on iNaturalist, nearly producing a false first record of a red-winged blackbird in Brazil before it was caught and removed.
The stakes are high because modern biodiversity research depends on these platforms. Population trend studies, conservation planning, and geographic distribution analyses all draw from citizen science records. Fabricated observations embedded in that data could skew findings in ways that are difficult to detect and costly to correct. Anderson fears the contamination will eventually reach scientific literature itself.
The phenomenon has historical precedent. In the early 1700s, a butterfly named the Charlton brimstone was formally described by Linnaeus and housed in the British Museum for decades before a specialist discovered that its distinctive blue markings had been painted onto the wings of a common brimstone. The hoax had embedded itself into taxonomy. Generative AI is that same impulse operating at industrial scale — the tools are free, fast, and accessible to anyone.
Yet the platforms are not without defenses. iNaturalist, which holds some 500 million images, has introduced community tools to flag AI-generated content. Experts advocate for image-authentication protocols, metadata verification, and public education. There is also a productive irony: AI, when constrained by human oversight, is among conservation science's most valuable instruments. Machine learning models have predicted bird migration weeks in advance, monitored algal blooms, and helped identify wildlife in camera trap images across Australia — all while keeping classification in human hands.
The distinction matters. As taxonomy faces an expert shortage and ecological crises accelerate, the best defense against AI-generated fakery turns out to be the same quality that made citizen science powerful in the first place: people who look closely, question what they see, and refuse to let the record be falsified.
In the leaf litter of Indonesian rainforests, a dead leaf mantis clings to a branch with the kind of perfection that evolution spent millennia engineering. Its body mimics the veins of dying vegetation so precisely that predators pass it by. But when Kris Anderson, a mantis specialist who has spent decades identifying insects on the citizen science platform iNaturalist, scrolled through a user-submitted image of what was claimed to be a Deroplatys from a Malaysian tropical forest, something felt wrong. The head shape was unfamiliar. The antennae didn't match anything in his mental catalog. The patterns across its body belonged to no species he had ever encountered. Anderson, who has made more than 26,000 identifications on iNaturalist, recognized the truth: this mantis had never existed. It was born from code, not evolution—a creature assembled by generative AI image tools.
Anderson is not alone in noticing the problem. Across citizen science platforms like iNaturalist and eBird, which collectively house hundreds of millions of observations that scientists rely on for research ranging from bird migration to conservation assessments, fake species are beginning to accumulate. In November 2025, a photograph of a willow tit—a palm-sized bird with black head feathers—appeared on a Scottish birding site, its beak inserted into a feeder. Alexander Lees, a reader in biodiversity at Manchester Metropolitan University, examined the image and felt a cold sweat. The bird's feet and beak were deformed in ways no living creature would tolerate. The image was entirely fabricated. Lees has been documenting an accelerating pattern: AI-generated fakes appearing on the very platforms that have become essential infrastructure for modern biodiversity research.
The threat extends beyond wholly invented creatures. In early July, Lees and colleagues published an opinion piece in Nature Ecology & Evolution warning that citizen science databases could become "contaminated" by AI-generated fabrication. They demonstrated the ease of the problem by asking Google's Gemini chatbot to generate a red-winged blackbird on a branch against a blue sky. The result was nearly convincing, marred only by an odd surplus of tail feathers. They then took a real photograph of a South American oriole and asked the AI to "make this look better." The system added a red flare to the bird's wing—a subtle enhancement that could fool an expert ornithologist or a computer-vision classifier into believing it was a different species entirely. This exact scenario has already occurred on iNaturalist: a user innocently touched up an image in this manner, leading to the first recorded sighting of a red-winged blackbird in Brazil. The image was detected and hidden before it could contaminate the scientific record.
Anderson worries that the problem will eventually reach scientific literature itself. He has written twice about fake mantises in the past six months, each time prompted by entries on iNaturalist. The concern is not merely academic. Because modern biodiversity research increasingly depends on citizen science records—for environmental assessments, conservation planning, species surveys—the injection of AI-generated observations could skew findings in ways that are difficult to detect. If a researcher pulls data from iNaturalist to study population trends or geographic distribution, and that data includes fabricated sightings, the conclusions built upon it become unreliable.
This is not the first time the scientific record has been contaminated by false species. In the early 1700s, a London chemist named James Petiver was gifted a spectacular butterfly with vibrant yellow wings marked by two bold blue moons. He named it the Charlton brimstone. By 1763, it had been formally described in scientific literature by Carl Linnaeus himself, the father of modern biological classification. It remained in the British Museum for decades until a Danish insect specialist named Johann Christian Fabricius examined it closely and discovered the truth: the blue moons were not marks of evolution but of human hands. Someone had painted them onto the wings of a common brimstone butterfly. Whether deliberate deception or elaborate prank, the hoax had embedded itself into taxonomy. A curator at the British Museum was reportedly so incensed that he stamped the specimen to dust.
Generative AI is a force multiplier for such deceptions. The tools are freely available, accessible to anyone with time and intent. On social media, fake species proliferate accompanied by AI-generated descriptions of their behaviors and habitats, often met with credulous belief. The Atlas of Living Australia, a digital infrastructure that aggregates biodiversity data across the country, has fielded submissions of unicorns, tyrannosaurs, and anonymous lovers—obvious fakes filtered out by vigilant curators. But as AI image generation improves, the line between obvious fabrication and imperceptible forgery will blur.
Yet citizen science platforms are not defenseless. They are populated by thousands of committed contributors and expert reviewers who donate vast amounts of time to maintain database integrity. iNaturalist, which houses some 500 million images, has recently added community tools designed to flag AI-generated content. Alexander Lees suggests that platforms need robust image-authentication protocols and metadata verification to ensure uploaded imagery represents legitimate observations. Education matters too: if the public understands the pitfalls of generative AI and citizen science systems, they may be more cautious about what they submit and believe. There is also a paradox worth noting. Machine learning and AI, when properly constrained and verified by humans, can serve conservation. An AI model trained on eBird data predicted bird migration patterns weeks in advance. The CitClops project used AI to monitor early signs of algal blooms in seawater. In Australia, machine learning systems trained on more than 1.1 million human-verified wildlife camera trap images now help conservationists identify emus and wallabies, reducing manual labor while keeping species classification in human hands. The difference is crucial: these systems are tools that assist human judgment, not replace it.
As climate change and biodiversity loss accelerate, and as the field of taxonomy itself faces a shortage of experts to describe and catalog species, the arrival of generative AI presents a triple crisis. But so far, the best defense against the flood of AI-generated fakery has proven to be the very thing that makes citizen science powerful: passionate, dedicated humans who care enough to look closely, to question what they see, and to protect the integrity of the record.
Citas Notables
Because so much modern biodiversity research increasingly incorporates citizen science records, I felt it was important to document this vulnerability early, before AI-generated observations became commonplace.— Kris Anderson, mantis specialist
Problems with denoising and 'beautifying' existing imagery are likely much more common than people trying to pass off entirely AI-generated imagery as a biological record, but both are highly problematic.— Alexander Lees, Manchester Metropolitan University
Quieres la nota completa?
Lee el original en Scitech ↗
Technology