Generative editing can alter a bird's diagnostic markings and apparent location, turning a photograph uploaded for identification into a bad research record; the Guardian describes an epaulet oriole photographed in Brazil that became an apparent red-winged blackbird after software was asked to improve the image. [1]

That mistake matters because citizen-science repositories help researchers map where species appear and how ranges change, so an attractive correction to a branch or feather can quietly change the observation itself rather than merely polish its presentation, a value that makes careful review more necessary rather than less. [1]

iNaturalist had flagged 1,400 images for AI use among more than 610 million, but those flags are reports from a detection process, not a random audit, a contamination rate or evidence that every flagged image altered a species or location record, and scientists say the undetected share is unknown.

The unknown share of undetected edits therefore prevents both reassuring claims that the problem is negligible and alarming claims that the platform has suffered a measured epidemic; outright hoaxes, ordinary edits and accidental alterations to diagnostic features also require different treatment.

The Nature commentary's useful contribution is narrower and more important: once a public photograph can feed scientific work, provenance and review become part of the data, while one documented false sighting establishes a vulnerability rather than the prevalence of corrupted records, and the next useful evidence would be platform controls and a blinded sample.

-- KENJI NAKAMURA, Tokyo