AI-Generated Images Slip Into Scientific Journals, Eroding Trust

When NASA's Artemis II crew captured a stunning view of Earth from lunar orbit in April 2026, the image resonated with the same emotional power as the iconic Apollo 8 "Earthrise" photograph from 1968. But in an era when anyone can generate a visually indistinguishable fake in seconds, that emotional connection now comes with a question: How do we know it's real?

The answer matters more than ever for science. AI-generated images are infiltrating peer-reviewed journals, and the traditional markers of visual credibility—technical sophistication, institutional backing, production difficulty—are losing their power to signal authenticity.

Fabricated Visuals Already Passing Peer Review

The problem has moved beyond theory. In 2024, two papers were retracted after publishing AI-generated figures showing biologically impossible structures. In April 2026, the New England Journal of Medicine retracted a paper containing a clinical image manipulated with AI, according to research published by Nan Li, an associate professor of science communication at the University of Wisconsin-Madison.

These high-profile cases likely represent a fraction of the issue. Researchers working in materials science and other visually dependent fields have warned that AI-generated images pose growing threats to evidence integrity. Academic publishers are deploying detection tools, but those systems consistently lag behind the generative models they're designed to catch. Detectors can only identify patterns they've been trained to recognize, requiring constant updates as new AI tools emerge.

The most dangerous fakes aren't obvious fabrications—they're realistic images that subtly distort scientific details while remaining plausible enough to pass initial review.

The Collapse of Visual Authority

For decades, scientific images carried inherent authority because they were difficult to produce. Microscope images, climate visualizations, and space photography required expensive equipment, institutional resources, and specialized expertise. That barrier to entry served as an implicit authentication mechanism.

Generative AI has eliminated that barrier. Anyone can now create polished, scientific-looking images from text prompts. When visual quality and institutional attribution become unreliable signals, audiences fall back on a less reliable heuristic: their existing beliefs.

This shift enables motivated reasoning at scale. Authentic scientific images that challenge someone's worldview can be dismissed as AI-generated, while fabricated images that confirm their beliefs are accepted as evidence. Science loses one of its most powerful tools for public communication when visual evidence becomes universally suspect.

Why It Matters

The credibility crisis extends beyond misinformation. When audiences stop trusting visual evidence altogether, scientists lose a critical bridge to public understanding. Images don't just illustrate findings—they create emotional connections and establish credibility for the underlying research. If every image can be questioned and none carries inherent authority, science communication faces a fundamental breakdown.

Transparency as a Path Forward

Li's research suggests that disclosure may help rebuild trust. Her team found that audiences familiar with AI tools viewed transparency about AI use as a positive signal—some even rated clearly labeled AI-generated content as more credible than unlabeled material.

The solution may lie in treating image provenance with the same rigor scientists already apply to data provenance. Researchers routinely disclose funding sources, methodologies, and conflicts of interest. Similar standards for visual evidence could specify whether AI was used to generate or modify an image, clarify whether it represents direct observation or illustration, and document verification methods.

What made the original Earthrise photograph meaningful wasn't just its beauty—it was the traceable connection to physical reality. Astronauts, cameras, documented missions, and verifiable observations stood behind the image. In the age of generative AI, that chain of custody can no longer be assumed. It must be explicitly documented.

The details were first reported by Nan Li in The Conversation, drawing on her research into visual science communication and public trust.