AI-altered bird photos threaten citizen science databases used in research

For birdwatchers, recording a species outside its known range has long been a prized discovery, often making headlines—as when a western reef heron, typically found in Africa and southern Europe, was spotted in north Wales in June and celebrated on birding forums. But this tradition is now threatened by the rise of AI-generated and AI-enhanced images flooding citizen science platforms. Scientists have issued a formal appeal in a Nature journal commentary, warning that the credibility of platforms like iNaturalist and Macaulay Library—which are routinely used in academic research to monitor species' habitat ranges and understand how wildlife responds to climate change—is at risk. On iNaturalist alone, which hosts over 610 million images, just 1,400 have been flagged for AI use, but researchers acknowledge this represents only the tip of the iceberg; many AI-altered images go undetected entirely. The problem is not primarily malicious. Rather, wildlife photographers, eager to create beautiful images, use generative AI platforms to enhance their pictures—removing obscuring branches, improving lighting, or otherwise polishing the composition. When they do, the AI algorithm can inadvertently introduce features from other bird species, creating hybrid images that look convincing but are scientifically false. Dr Alexander Lees, an ecologist at Manchester Metropolitan University and author of the Nature commentary, pointed to the example of a red-winged blackbird reported in central Brazil, an area where this North American species had never been recorded. The bird turned out to be an epaulet oriole—a common neotropical species—that a photographer had submitted to the AI to "make it look better." The algorithm added red-winged blackbird features to the image, generating a false sighting. Lees noted that while outright hoaxes remain rare and easy to identify (nobody will believe a toucan in Siberia), these subtle AI-introduced alterations are far harder to catch. Tony Iwane, iNaturalist's director of community support and a co-author on the Nature paper, acknowledged that most AI-contaminated submissions are unlikely to be intentionally deceptive, but he appealed to users to be vigilant. He emphasized that regular citizens posting on platforms like iNaturalist provide data that scientists could never collect at scale—from monitoring whether plants flower earlier as the climate warms to tracking whether species migrate northward—and that this real-time global sensor of nature's changes depends entirely on accuracy. The more precisely scientists know where species are located, the better informed conservation decisions can be, but contaminated data undermines that mission.

The rise of generative AI platforms such as ChatGPT and Google Gemini has created an unprecedented challenge for citizen science: the same tools that make photo editing accessible to everyone have made it nearly impossible for scientists to distinguish genuine observations from AI artifacts. While outright hoaxes—such as a toucan sighting in Siberia—remain rare and easy to spot, the more insidious problem is that ordinary birdwatchers, motivated by a desire to capture beautiful photographs, are unknowingly corrupting the scientific record. Dr Alexander Lees's example of the red-winged blackbird in Brazil illustrates the mechanism: a photographer enhanced a common epaulet oriole using AI, and the algorithm blended in features from a rarer species, creating a false sighting that could mislead researchers studying species range expansion and climate-driven migration. The platforms themselves acknowledge the threat but have limited detection capacity—iNaturalist's 1,400 flagged images represent less than 0.002% of its database, and researchers admit that the true prevalence of AI contamination remains unknown. This asymmetry between detection capability and the ease of creating high-quality fake images suggests the problem will likely grow as AI tools become more sophisticated and widely used.