On Saturday, I saw a post by Rahul Ganatra on Roon about NEJM retracting their Image in Clinical Medicine due to it being altered with AI.
I immediately jumped over to Twitter to post this juicy tidbit and it took off.
200k impressions in three days. That’s the most impressions I have had for a tweet in the last year (granted, I haven’t been posting so much at Twitter in the last year, and X Analytics don’t allow me to go further back in time).
Here is the text of the retraction
To the Editor: We were unaware of Journal policies on image manipulation and had altered our submission by using an artificial intelligence (AI) tool to move the ruler to the top of the image. We therefore wish to retract our image and case report.
Yuling Wang, M.Med.
Xiangdong Mu, M.D.of the week
This is the first retraction by the New England Journal of Medicine since 2020, when it was burned by the Surgisphere scandal. As always, Retraction Watch has excellent coverage.
Why is this such a delicious story?
Part of it is the spectacle of the mighty falling. We imagine The NEJM as impenetrable guardians of rigor, immune to the misinformation that plagues the rest of us. So when they get fooled, it’s comforting. If they can be duped, maybe the rest of us aren’t so foolish after all.
But there’s something deeper going on here.
Most of what appears in NEJM is armored with methodology, prospective design, pre-specified analyses, adjudicated outcomes. Even when wrong, it is wrong in a structured, auditable way.
But Images in Clinical Medicine? It’s just a picture, a paragraph, and the implicit contract that seeing is believing.
And generative AI has rendered that contract null and void.
The retraction forces us to confront something uncomfortable: images are no longer self-authenticating. The assumption that a clinical photograph represents reality no longer holds.
A convincing image no longer requires a patient, a camera, and a disease. It just requires a prompt.
That shifts the burden. The whole premise of Images in Clinical Medicine is: look at this.
But now the first question isn’t “What is it?”
It’s “Is it even real?”
Ronald Reagan famously said we need to “Trust but verify.”
Today, in the NEJM, it’s just “Verify.”
