AI or Not Achieves 100% Detection of Deepfake X-Rays in Independent Test

AI or Not announced a successful independent benchmark test demonstrating 100% detection of deepfake X-rays using a curated dataset from a recent Radiology study. The technology achieved 95% overall accuracy, outperforming board-certified radiologists who identified only 41% of deepfake images initially, improving to 75% when informed of the presence of AI-generated content. Four leading multimodal large language models (LLMs) showed accuracy ranging from 57% to 85%. The study highlights the vulnerability of medical imaging to generative AI and emphasizes the need for a multi-faceted approach to combat this issue, including clinician training and watermarking. AI or Not's platform also reported a 7.8% false positive rate on authentic X-rays, indicating its reliability in distinguishing real from synthetic images. The findings point to significant implications for patient safety and insurance claims in the medical field.

Key Points: • AI or Not achieved 100% detection of synthetic X-rays in a recent benchmark test. • Board-certified radiologists identified only 41% of deepfake X-rays initially. • The study emphasizes the need for combined efforts in detection, training, and governance.