Scientific images—once the gold standard of evidence—are now under siege from AI-generated fakes that can fool even peer-reviewed journals, raising urgent questions about how the public will ever trust visual proof again. Since 2023, at least three high-profile papers have been retracted after AI-manipulated figures were discovered, including a 2024 study in Nature that featured biologically implausible cell structures and a 2026 clinical image in the New England Journal of Medicine later revealed to have been altered with generative AI tools [Nature]. The problem isn’t just misinformation—it’s a systemic erosion of trust in the very visuals that underpin scientific authority.
For decades, scientific images carried weight because they were difficult to produce. Microscope slides, satellite photographs, and X-ray scans required expensive equipment, institutional backing, and specialized expertise. The assumption was simple: if an image came from a lab or space agency, it was real. But today, anyone with a text prompt can generate a convincing fake in seconds. A 2023 study in Science Advances found that 60% of participants struggled to distinguish AI-generated medical scans from real ones, even when prompted to look for inconsistencies [Science Advances]. The result? A crisis of credibility that threatens to unravel one of science’s most powerful tools for public communication.
This shift isn’t just theoretical. In fields like materials science, where visual evidence is critical, researchers warn that subtle AI distortions—ones that pass initial peer review—could lead to flawed conclusions being accepted as fact. Meanwhile, publishers are scrambling to adopt AI-detection tools, but experts say these systems will always play catch-up to the rapidly evolving generative models. The bigger risk? That audiences, disillusioned by the flood of AI fakes, will dismiss all scientific images—whether real or not—as unreliable.
How AI Is Infiltrating Science—and Why Detection Tools Aren’t Enough
Generative AI tools like MidJourney, DALL·E, and Stable Diffusion have become staples in research labs, used to create illustrations, simulate data, and even enhance real images for clarity. The benefits are clear: scientists can now visualize complex concepts more quickly and accessibly. But the tools also blur the line between illustration, enhancement, and outright fabrication.
Take the case of a 2024 paper published in Cell Reports that was retracted after reviewers flagged AI-generated figures depicting protein structures. The images looked plausible at first glance, but closer inspection revealed biological impossibilities—like proteins arranged in ways no cell could naturally produce. The authors admitted they had used AI to “improve” the visuals, but the lack of disclosure meant the manipulation went unnoticed until after publication [Cell Reports].
The problem extends beyond retraction-worthy mistakes. In April 2026, the New England Journal of Medicine retracted a clinical study after an anonymous tipster pointed out that a key diagnostic image had been manipulated with AI. The journal’s editors noted that while the alterations were subtle, they “compromised the integrity of the visual evidence” [NEJM]. Such cases are likely just the tip of the iceberg, with researchers in fields like materials science warning that AI-generated visuals pose growing threats where visual evidence is paramount.
Publishers are responding with AI-detection tools, but these systems face an uphill battle. Most detectors rely on pattern recognition—identifying artifacts or inconsistencies in pixel data. Yet as AI models improve, they produce images that are increasingly indistinguishable from real ones. A 2023 report from the Journal of Medical Imaging found that even state-of-the-art detectors missed 30% of AI-generated medical images when tested against a dataset of real scans [JMI]. The bigger concern? Subtle distortions that don’t trigger detection flags but still skew scientific conclusions.
Why Trust in Scientific Images Is Crumbling—and What’s at Stake
For centuries, scientific images carried authority because they were hard to fake. A photograph of Earthrise from the Apollo 8 mission in 1968 wasn’t just a picture—it was proof of human achievement, backed by astronauts, cameras, and verifiable missions. Today, that traceable connection is eroding. With AI, anyone can generate a visually identical “Earthrise” image in seconds, complete with craters and atmospheric glow. The question isn’t just whether people can tell the difference—it’s whether they’ll believe any image at all.
Research in science communication shows that people rely on three key shortcuts to judge visual credibility:
- Technical sophistication: Does the image look like it came from a high-end lab or telescope?
- Institutional trust: Is it affiliated with a reputable organization?
- Confirmation bias: Does it align with what I already believe?
AI undermines all three. A polished, AI-generated image can look as technically sophisticated as a real one. Institutional attribution becomes meaningless when images circulate online without context. And confirmation bias kicks in: people accept AI fakes that align with their views while dismissing real images that challenge them.
This dynamic amplifies what psychologists call “motivated reasoning”—the tendency to accept evidence that supports preexisting beliefs and reject evidence that contradicts them. The result? Authentic scientific images that challenge public narratives (like climate data or vaccine efficacy) are increasingly met with skepticism, while fabricated images that confirm biases are accepted as truth. If audiences stop trusting visual evidence altogether, science loses one of its most powerful tools for public communication.
Key Takeaway: The crisis isn’t just about fake images—it’s about the collapse of the heuristics that once made scientific visuals credible. Without new standards, trust in science’s visual evidence could unravel entirely.
Transparency Over Restriction: The Path Forward
AI tools aren’t going away, and neither is their role in scientific communication. The challenge is using them without transferring AI’s credibility deficit onto the science itself. Experts agree: the solution lies in transparency—treating image provenance with the same rigor scientists already apply to data.
Just as researchers disclose funding sources, methodologies, and conflicts of interest, they should now clarify how images were created. Was AI used to generate or modify the visual? Is it a direct observation, a simulation, or an illustration? What does it represent, and how was it verified? Can it be replicated?
A 2023 study published in PLOS ONE found that participants rated AI-generated content as more credible when it was clearly labeled—even if they were skeptical of AI tools in general. The key is giving audiences the context to evaluate what they’re seeing [PLOS ONE]. Transparency alone won’t resolve every dispute, but it’s a critical first step.
Professional societies are beginning to develop standards. The American Society for Cell Biology now requires authors to disclose AI use in figure legends, while the Nature Portfolio has introduced guidelines for labeling AI-generated images. Yet the field lacks unified standards, leaving room for inconsistency. Some researchers argue for outright bans on AI in certain types of visual evidence, while others advocate for mandatory disclosure and third-party verification.
What’s clear is that scientific institutions can no longer assume audiences will automatically trust their visuals. Trust now depends on documentation, clear communication, and collective adherence to evidence-based norms. Without these, science risks entering a world where every image can be questioned—and no image carries inherent credibility.
Why Authentic Images Still Matter—and How to Protect Them
The original Apollo 8 “Earthrise” photograph isn’t just iconic—it’s meaningful because of its traceable connection to reality. Behind every frame were astronauts, cameras, and documented missions. That authenticity is what gives scientific images their power. In the age of AI, that connection must be explicit.
So how can researchers, publishers, and the public navigate this new landscape? Experts offer three immediate actions:
- Mandate disclosure: Require clear labeling of AI-generated or modified images in research papers, with details on how the image was created and verified.
- Adopt verification protocols: Develop field-specific standards for validating visual evidence, including third-party checks for high-stakes images.
- Educate audiences: Public campaigns should teach critical thinking about visual evidence, emphasizing the importance of provenance and transparency.
The stakes are high. If trust in scientific images erodes, so too does the public’s faith in the institutions that produce them. The alternative—a world where every image is questioned and none carries inherent credibility—is one science can’t afford.
What Happens Next: The Road Ahead
The next critical checkpoint will be the 2024–2025 updates to journal guidelines from major publishers like Nature, Science, and the NEJM, expected to formalize AI disclosure requirements. Meanwhile, the European Commission is drafting regulations on AI-generated content in scientific publishing, with proposals likely to be finalized by mid-2025 [EU AI Act]. Researchers and institutions are also pushing for standardized metadata tags to track image provenance, similar to how data repositories now log experimental conditions.
For now, the burden falls on individual scientists to adopt transparency. As one materials scientist told Science in 2023, “The bar for visual evidence has never been higher—and the consequences of failing to meet it have never been more severe” [Science]. The question is whether the scientific community can rise to the challenge before trust in its visuals is lost forever.
What you can do: Stay informed about emerging guidelines from your field’s professional societies. If you’re a researcher, start disclosing AI use in your visuals—even if not yet required. For the public, ask questions: Where did this image come from? How was it created? And most importantly, can it be trusted?
This article is part of World Today Journal’s ongoing coverage of AI’s impact on scientific integrity. For updates on evolving standards and case studies, subscribe to our Science & Technology newsletter.
