Imagine a world where a simple X-ray could be faked so convincingly that even the most trained doctors can't tell it apart from a real one.

This is no longer science fiction—AI-generated "deepfake" X-rays are now so realistic that they are fooling radiologists and even AI models. A recent study published in Radiology, the journal of the Radiological Society of North America, on March 24, 2026, explores this emerging issue and raises concerns about the potential misuse of these synthetic images in the medical field.

The Risk of Synthetic Medical Images

A "deepfake" refers to media—photos, videos, or in this case, X-rays—that appear authentic but are artificially created or altered using AI. While deepfakes are commonly associated with video and audio manipulation, the new challenge is in the medical imaging world, where AI is generating X-rays that are nearly indistinguishable from real ones. This opens the door for fraudulent medical claims, tampered diagnoses, and even legal disputes if a fake image can be passed off as authentic.

According to Dr. Mickael Tordjman, lead author of the study and a post-doctoral fellow at the Icahn School of Medicine at Mount Sinai, the risk is high. "These deepfake X-rays are realistic enough to deceive even radiologists, the most highly trained image specialists," he explained. "This creates a major vulnerability in legal cases, where a fabricated fracture could easily be mistaken for a real one."

How the Study Was Conducted

The research involved 17 radiologists from 12 institutions across six countries, including the United States, France, Germany, Turkey, the United Kingdom, and the United Arab Emirates. The study analyzed 264 X-ray images—half of which were real, and half generated by AI models like ChatGPT and RoentGen, an open-source AI model developed by Stanford Medicine.

Radiologists were shown two sets of images. The first set contained a mix of real X-rays and ChatGPT-generated X-rays, while the second set focused exclusively on chest X-rays, with half generated by RoentGen. The results were shocking: when radiologists were unaware that fake images were included, they only correctly identified 41% of the AI-generated X-rays. When they knew fake images were involved, accuracy rose to 75%.

How AI and Radiologists Struggle to Detect Deepfakes

The study revealed that even the most experienced radiologists struggled to distinguish fake X-rays from real ones. The performance varied from 58% to 92% in recognizing ChatGPT-generated images. Four AI models—GPT-4o, GPT-5, Gemini 2.5 Pro, and Llama 4 Maverick (Meta)—performed similarly, with accuracy ranging from 57% to 85%. Even the ChatGPT-4o model, which was responsible for generating the deepfakes, failed to detect all of the images, though it did outperform the other models.

Visual Clues in Deepfake X-Rays

Despite the impressive quality of these synthetic X-rays, there are certain visual patterns that can give them away. Dr. Tordjman points out that deepfake medical images often "look too perfect." For example, bones may appear overly smooth, spines may be unnaturally straight, lungs might be too symmetrical, and blood vessels may look excessively uniform. Additionally, fractures often appear too clean and consistent, sometimes affecting only one side of the fragment.

Potential Dangers and the Need for Safeguards

The findings of this study highlight the serious risks posed by deepfake X-rays in the medical field. These fabricated images could easily be used in legal disputes or to manipulate diagnoses, potentially putting patients at risk. Furthermore, if hackers gain access to hospital systems, they could inject synthetic images into patient records, causing widespread clinical disruption.

To combat these threats, the researchers propose stronger digital safeguards. These could include embedding invisible watermarks or cryptographic signatures directly into images, allowing for the verification of authenticity and the identification of the technologist responsible for the image capture.

The Future of AI in Medical Imaging

“We are potentially only seeing the tip of the iceberg,” says Dr. Tordjman. As AI technology advances, the next logical step could be the creation of synthetic 3D medical images, such as CT and MRI scans. Establishing detection tools and educational datasets now is critical to ensuring that healthcare professionals are equipped to recognize deepfakes as they become more sophisticated.

To help train radiologists and other healthcare professionals, the research team has released a curated deepfake dataset along with interactive quizzes designed to raise awareness and improve detection skills.

Final Thoughts: A Call to Action

The increasing sophistication of AI-generated medical images presents a significant challenge for the healthcare sector. As these images become more convincing, the potential for misuse grows. It's clear that as AI continues to develop, stronger safeguards, enhanced training, and better detection tools will be essential to protect the integrity of medical imaging and prevent the manipulation of patient data.

This study serves as a wake-up call—underscoring the need for constant vigilance as we embrace the power of AI in medicine. As the technology evolves, we must ensure that our ability to detect and safeguard against deepfakes evolves just as quickly.