AI or Not said its detection product scored 95% overall accuracy and 100% detection of synthetic X-rays in a blinded independent benchmark on a curated dataset drawn from Mount Sinai research on deepfake medical imaging, according to a company release on PR Newswire dated 12 May 2026.
On authentic X-rays in the same curated set, the company said 92.21% were correctly cleared as real, implying a 7.8% false-positive rate, with a 0% miss rate on AI-generated images. The underlying Radiology study context cited radiologist accuracy of about 41% without warning rising to roughly 75% when alerted, and multimodal LLMs in the 57–85% range — those figures are study-reported, not AI or Not's own radiologist trial. The company said the curated set's generator was GPT-4o and framed the images as out-of-distribution relative to its training data.
Buyer angles named in the release include medical imaging integrity and insurance claims verification. Accuracy percentages and comparison claims are vendor-stated benchmark results and have not been independently verified by AI-360.