Nikon disqualifies its Small World winner over generative AI

"Enhanced" stopped being a safe word this week.

Nikon has pulled first place from the winner of its Small World in Motion microscopy video contest for breaking the competition's rules on generative AI.

Nikon reversed its own judges

In a statement dated Oct. 9, Nikon says it re-evaluated the 2026 first-place video, consulted its judging panel and found it "did not comply with the competition rules regarding generative AI." It has adjusted the rankings.

Nikon says the decision concerns only the video's eligibility, not the entrant's reputation, scientific work or intent. It also says it will revisit its rules and evaluation procedures for future entries.

The winning video, by Ning Xu, was presented as showing abnormal cilia beating in the airway of a child with primary ciliary dyskinesia, according to The Scientist.

Weeks earlier, The Scientist reports, Nikon and Xu had said AI was used only to enhance the images, which the rules allowed. Nikon called it "a tool for enhancement of microscopy-generated source data, not for image generation from scratch."

Scientists disagreed. Robert Hirst, a lead scientist at the University of Leicester, said the cilia "look nothing like those from PCD patients," according to The Scientist.

First place now goes to Nguyen Nam Nhat, for a video of a tiny roundworm and a single-celled Dileptus exploring their microscopic world, Nikon says.

Enhancement and generation are the same tools

This is a small contest with a big precedent.

The rules drew a line between AI that cleans up real data and AI that makes things up. The trouble is that many enhancement tools are generative models underneath. They fill in detail the sensor never captured. Push them hard enough and they invent structure, and a judge looking at the output cannot always tell which happened.

Nikon's first answer trusted the entrant's description of the process. Its second answer came after outside experts looked at the result and said the biology was wrong.

So the result, not the label on the tool, decided it.

That should worry anyone who runs microscope images, satellite photos or product shots through an AI upscaler and calls the output real.

Keep the receipts

Detection is getting easier, too. Google opened its SynthID Detector to the public this week, and AI watermarking means some generated pixels now carry a signal anyone with the right tool can check.

If AI is anywhere in your pipeline for images that claim to show something real:

1. Keep the raw, unprocessed files, with their original metadata.
2. Write down every AI step: the tool, the model, the settings.
3. Disclose it up front, in the caption or the methods section, before anyone asks.
4. Read the rules of the contest, journal or client you are submitting to. If they are vague, ask in writing.

If you cannot show the before, you cannot defend the after.

Watch the new rules

Nikon says it will revisit its rules and evaluation process for future entries. Watch what it writes, because other photo contests and journals will copy the wording.

Save your raw files today.