Nature warned on July 22: scientists are already slipping AI-generated images into research papers, and journals have no clear rules on what's allowed. A telling case - a biologist produced convincing but fake lab photos with a single ChatGPT prompt. The line between an image for illustration and an image as evidence has become a question of trust in science itself.
- On July 22, 2026, Nature published a piece on the risks of AI-generated images in scientific publications.
- Example: a biologist published convincing fake western blot images, made with a single prompt to ChatGPT.
- The problem: journals offer little guidance; the ethical line is whether the image is evidence, or just illustration.
I've spent my whole life looking at form, and I'll say it straight. The most dangerous fake photo isn't the crude one. The smooth one is dangerous - the one that looks exactly right and doesn't give you a single edge to grab onto.
Here's what stops me. Fakery in science isn't new - people have been cutting and pasting photos for years. What's new is the price. Until recently, a good fake took skill and hours; now it takes just one sentence and ten seconds.
And this isn't really about the technology. It's about trust. Science rests on one quiet assumption - that the photo in the paper shows what actually happened in the lab. Pull that assumption out, and every photo needs a second check. And a second check slows down everything and everyone.
And don't think this is only a problem for scientists. You also take 'the photo' as proof - in the news, in the ad, in the online argument. The conclusion is simple and uncomfortable: a convincing picture no longer proves anything on its own. Ask where it came from, the way we've learned to ask where the text came from. The eye alone isn't enough anymore.