A Nikon Microscopy Contest Winner May Be AI-Generated. Scientists Spotted It in Days.

A video that won the 16th annual Nikon Small World in Motion contest is under re-review after a UT Southwestern PhD student spotted a Google DeepMind SynthID watermark in the file and a researcher said the purple structures visible alongside the main subject "do not occur in biology."

The winning entry, by optical engineer Ning Xu of the National University of Singapore, purports to show cilia, the hair-like structures that move fluid across cell surfaces, propelling themselves inside lung tissue from a child with primary ciliary dyskinesia, a rare inherited condition that disrupts that motion. Nikon Instruments announced the result in mid-September 2026 and reposted the video to LinkedIn in late September. Within days, working scientists had identified specific anomalies that, taken together, point toward generative AI fabrication rather than a microscope capture.

Ian Donovan, a PhD student at UT Southwestern Medical Center, told CNN he had identified a SynthID watermark in the file. SynthID is a Google DeepMind technology that embeds an invisible marker in AI-generated pixels so they can be traced back to their source. The watermark is the closest thing to a smoking gun the public discussion has produced, because it does not require expert interpretation of the image; it is metadata carried with the pixels themselves.

The structural objection came from Edward Phelps, who wrote on LinkedIn that the red, blue, and purple sub-epithelial structures in the video are biologically implausible. "Such a sub-epithelial structure composed of extracellular mitochondria of the same size as cell nuclei does not occur in biology," Phelps said of the purple clusters, and he added that the blue nodules resembled nuclei but did not behave like nuclei under the conditions shown. The cilia themselves are consistent with real microscopy; the supporting anatomy is what cannot be.

The contest's own rules appear to rule out AI generation. Patrick Hickey, a fifth-place winner, told the BBC the rules were "quite clear" that entrants could not use generative AI and that submitted imagery had to be captured under a microscope. The official 2026 Nikon Small World in Motion rules page contains a comparable AI-generation prohibition, according to the same reporting. The contest has run for 47 years, and the rules were written long before generative image models could fabricate pixels outright.

Nikon Instruments has acknowledged the dispute. In LinkedIn updates reposted by the company, the sponsor said it is "carefully re-reviewing" the submission and that Xu is "respectfully complying" with that re-review. Xu has not publicly disputed the detection chain, and there is no source basis for asserting motive or intent. The outcome of the re-review, and any change to the contest's rankings, is pending.

The verification chain worked. Distributed experts, a public watermark, and a sponsor's own re-review process caught the suspected fabrication within days of the LinkedIn repost, not months. That pace reflects how the scientific community's verification culture is adapting to generative AI in real time: a PhD student in one city, a bioengineer in another, and a contest organizer on a third platform each contributed a different piece of the case, and the conversation moved fast enough that the sponsor had to acknowledge it before the next news cycle.

The rules, however, are still catching up. The widely cited 2010 scientific-imaging guidelines permitted stained or colorized images if the changes were disclosed and described. Those rules predate generative AI's ability to fabricate pixels outright, and the Nikon contest's rules were written for that older era. As AI image generation spreads, contest organizers, journals, and image-integrity services are all being pushed to update the rules and the watermarking norms that support them.

For now, the Small World in Motion winner remains a working example of the gap: a watermark and a structural objection, surfaced on LinkedIn and picked up by Nature, the BBC, and CNN, have given the sponsor a clear path to re-review. What happens next, and how the contest updates its rules for 2027, will tell readers how seriously the field treats the new technology.