The Threat of AI to Media Archives

The Macaulay Library and eBird, community-driven citizen science projects, are facing a surge in AI-manipulated and AI-generated photos that threaten the scientific value of their archives, according to a new publication in Nature. The projects have implemented a strict policy prohibiting AI-altered images, warning that repeated violations may result in accounts being set to Not Public.

Community-driven science projects such as eBird, the Macaulay Library, and iNaturalist are seeing a rapid uptick in AI-manipulated and AI-generated photos. A new publication in Nature https://rdcu.be/ftx6h has recently highlighted the threat of AI photos to these projects, leading to some popular news articles https://www.theguardian.com/environment/2026/jul/20/ai-slop-manipulated-fake-images-birds-citizen-science-aoe . We want to raise awareness and inform the community of how AI reduces the scientific value of photos. The Problem AI is increasingly being pushed into our daily lives—nearly every updated piece of technology and your favorite applications have added various forms of AI, some optional and some unavoidable. Use of AI can add or remove details from images, changing your image into something that does not represent what the bird actually looked like while birding. This can be extreme, as in the case of the Green Heron where a third leg was added, or be more subtle, such as creating false feather detail to make a picture appear sharper. Even more troubling, many people may not even realize the extent of changes that AI is making to their photos, since they may just appear “better”, clearer, and sharper at first glance. Furthermore, many phones and editing applications have changed to using AI for photo processing, often without your knowledge and with no option to disable it. The following example is from a recent Nature article https://www.nature.com/articles/s41559-026-03141-y.epdf?sharing token=JEu 0kUalhm-oSGzZpOc4NRgN0jAjWel9jnR3ZoTv0MIXYSGF7IH-YSv2oP4xX-JIwr LTtxwIQ6LrDFLSyJ2 Eyv-ExAWdrUSm-v7qgMPul0 3OcbDpubaY965CMem4OU-QMBnLegaWuJA67Li19tYBnkdwnB0tpt71NKb Nyc%3D discussing this problem. When the original image of the Epaulet Oriole image B was edited by Gemini 3 to ‘improve’ its quality, it resulted in C an image that almost perfectly resembles a different extralimital species which is not in the same genus—the Red-winged Blackbird—and would be expected to fool both expert ornithologists and computer vision image classifiers. Why This Matters The Macaulay Library aims to document the natural world through digital media for scientific studies and future research and applications, and as such photos submitted to it through eBird are considered a record of where and when species occur, their behavior, what they look like, and the habitat they occur in. AI-introduced changes alter that natural truth, eroding the credibility of the ML archive among other birders and the scientific community. Policy Overview Use of AI to generate or alter photos submitted to eBird checklists and Macaulay Library is strictly prohibited. For photos you are submitting to eBird checklists, don’t use AI to process them and avoid using AI-powered tools within your preferred photo editing software. This might include buttons/tools like “Improve”, “Upgrade”, “Enhance”, or “Generate” in common photo editors. Also, be aware that some newer phones will automatically use AI to sharpen very zoomed-in photos. DON’T: - Don’t put AI-manipulated photos into eBird checklists. - Don’t use AI chatbots or generative AI tools to “enhance” or “improve” photos for your eBird checklists. - Don’t overuse AI-powered editing tools on photos you upload to eBird; avoid using them at all when possible. - Don’t use AI to generate images of what you thought you saw as documentation in an eBird checklist. DO: - Do keep your photos as natural as possible. - Do retain the original metadata for your photos when editing and exporting. - Do describe what you saw with words, especially if you didn’t get a photo. Read our full AI policy. Individuals repeatedly found to be uploading AI-generated or significantly altered images may have their accounts set to Not Public. What is eBird and the Macaulay Library doing to help? In the coming weeks, tags will be added to photos, video, and audio submitted to the Macaulay Library through eBird that are found to be AI-manipulated or over-edited. These tags will help eBird and Macaulay maintain the integrity of the archive and clarify our media-editing policy. Artificial tags will be added to media determined to be fully or substantially AI-generated, or edited so extensively that it no longer reliably depicts the organism, sound, or environment. Artificial media will be fully hidden from all public output. Over-edited tags will be added to media determined to be edited beyond the limits described in our guidelines https://support.ebird.org/en/support/solutions/articles/48001284842 but will remain publicly accessible, although they might be excluded from some research projects and might not be used to improve Merlin or other Lab tools. These tags will be applied as new media are found or brought to our attention. If you see these tags on your own media, you can 1 replace it with a less edited or non-AI version and/or 2 leave it be but change your editing workflow to avoid the same issue for your future uploads. What can you do to help? Please follow our best practices outlined above and be sure your photos accurately reflect what the bird looked like in the field. AI tools that generate content or change existing details are never acceptable. Even AI-powered editing tools like noise reduction and sharpening should be used only lightly, if at all. An image with artificial backgrounds or features might be nice to look at, but it is not scientific data and should not be uploaded to eBird and the Macaulay Library. In the future, the eBird community may also be able to flag this type of content directly. In the meantime, we encourage you to help identify media that is potentially artificial or over-edited by notifying us using this Google Form. Examples of AI photos in Macaulay Below we outline the main ways that AI usage can lead to wholly or partially falsified data using actual examples of photos submitted to Macaulay Library: 1. Over-editing Even without AI, photographers have long been able to manipulate their photos using various editing software. When editing, the goal should always be to have a natural reproduction of how the bird looked in life. Overuse of any editing tool, regardless of AI, can also result in very strange-looking birds. We ask eBirders to avoid over-saturation, over-sharpening, or augmenting the photo beyond what you observed. Recent advancements in AI tools in commonly used photo editing software take this to the extreme, such as AI sharpening and denoise tools adding fake “clarity” to bird feathers. Photographers should never turn any editing settings to the max, like shadows, sharpness, and brightness, as they are guaranteed to create an unnatural looking image. Even default levels of AI denoise are likely to introduce fake details. 2. “Enhancement” People seeking to “enhance” or “improve” their photographs using AI chatbots like Google Gemini are almost guaranteed to inadvertently remove natural details like insects and plants while adding fake details like new color patches or changes to beak morphology. For example, we’ve seen an uptick in people asking AI chatbots to “make this photo better” or similar prompts. Remember that these AI tools are a “black box”—you have no idea what they are doing to your photo and you have little to no control over it, so don’t use them for eBird photos Removing branches and using any type of AI generative fill to replace the branch or completely replacing the background of a photo also fabricates information and should never be done. It’s okay to have a branch in front of the bird At first glance the bottom picture above may appear identical and cleaner. But if you look closely, you’ll notice that Gemini changed critical biological information in the picture. Most importantly, the enhanced image removes the food the sparrow had in its bill which removes evidence of the type of food the sparrow eats and potential evidence of breeding at that location , generates an entirely fake lower bill which was covered by the food in the original image , and adds a dark malar line on the throat that Grasshopper Sparrows do not have. This illustrates the danger of using AI to enhance photos—it can fundamentally change the natural history, morphology, and plumage of the subject in the original photo. The original photo is great, and exactly what we would want added to the Macaulay Library without passing it through AI. 3. AI-generated While still rare, we have seen a few instances of fully generative AI photos meant to deceive others or to make a nicer representation of what the user saw. Never, under any circumstance, submit a fully generative AI photo to eBird. Intentional falsification is an egregious violation of our Community Guidelines https://support.ebird.org/en/support/solutions/articles/48001257989-ebird-and-merlin-community-guidelines and can result in your eBird account being permanently set to Not Public, removing all your checklists and media from public and scientific uses.