AI or Not says its detection software achieved 95% overall accuracy and caught 100% of synthetic X-rays in a blinded test on a dataset from a Radiology study, the company said on 12 May 2026.

The dataset came from “The Rise of Deepfake Medical Imaging”, a peer-reviewed study published in Radiology in March 2026 by researchers at the Icahn School of Medicine at Mount Sinai. AI or Not ran the test itself and said the images were not part of its training data.

On authentic X-rays, the software cleared 92.21% as real, a 7.8% false positive rate. In the original study, radiologists identified 41% of deepfake X-rays in an initial blinded review, rising to roughly 75% when told AI-generated images were present. Four multimodal LLMs ranged from 57% to 85% accuracy.

The study’s authors flagged risks including insurance fraud using fabricated imaging, tampered evidence in litigation and disability cases, and manipulated scans affecting treatment.

For insurers handling medical claims, the results are company-reported and point to synthetic imaging as an emerging fraud vector.


Share this post
The link has been copied!