The oldest argument in photography just got new rules, and they tell you exactly where the line is between capture and creation.
The Summary
- World Press Photo 2026 winner announced with strict AI usage rules that define what counts as a "photograph" in the age of generative models
- The winning image, "Separated by ICE" by Carol Guzy, shows children with their father after an immigration hearing and had to pass new AI authenticity requirements
- The contest draws a clear line: you can use AI for workflow, but not for reality itself
The Signal
World Press Photo just did what every institution dealing with AI will eventually have to do: define the boundaries between tool and creator. The 2026 competition rules allow AI for technical enhancement but ban it for generating, replacing, or adding content that wasn't in the original scene. This isn't philosophy. It's policy.
The distinction matters because photojournalism sits at the intersection of trust and technology. When Carol Guzy photographed those children clinging to their father, she captured a moment. If she'd used Midjourney to add a child who wasn't there, she would have created propaganda. The contest rules recognize this isn't a spectrum. It's a binary.
"The contest draws a line between using AI as a darkroom and using it as a fiction engine."
Here's what you can do under the new rules:
- Use AI for noise reduction, color correction, or technical cleanup
- Apply AI-powered editing tools that enhance what was already captured
- Employ computational photography features built into modern cameras
Here's what disqualifies you:
- Generate new elements that weren't in the scene
- Remove people, objects, or context from the frame
- Add content using AI synthesis, even if it "could have been there"
This framework works because it preserves the core contract of photojournalism: the photographer was there, this happened, you're seeing what they saw. Everything else is post-processing, a practice as old as the medium itself. Ansel Adams spent hours in the darkroom. The difference is he couldn't conjure a moon that wasn't in the sky.
The bigger implication runs beyond photography contests. Every field dealing with AI-assisted work is facing this same question: where does the tool end and the creation begin? Code written with Copilot suggestions is still code you wrote. A legal brief with AI research is still your argument. But a photograph with AI-generated subjects isn't a photograph. It's an illustration.
The Implication
Watch for more institutions to adopt frameworks like this. The World Press Photo rules work because they're verifiable and bright-line. You either captured it or you didn't. As AI tools become default features in professional software, knowing which side of that line you're on becomes the difference between documentation and fabrication. For anyone building AI tools for creative or professional work, this is your blueprint: enhance the human's capture, don't replace it.