Removing a watermark from an image leaves a digital scar that is far easier to detect than the original watermark.
April 29, 2026
Original Paper
The Forensic Cost of Watermark Removal
arXiv · 2604.25491
The Takeaway
Attackers often use specialized tools to scrub watermarks from AI-generated images to hide their origins. This study reveals that these removal tools leave behind distinct statistical artifacts in the image data. A simple classifier can detect these artifacts with extremely high precision, even if the image looks perfect to the human eye. This means the act of trying to hide an image's source actually creates a permanent record of the tampering. Digital forensics can now use these leaks to identify malicious actors who believe they have successfully cleared their tracks.
From the abstract
Current watermark removal methods are evaluated on two axes: attack success rate and perceptual quality. We show this is insufficient. While state-of-the-art attacks successfully degrade the watermark signal without visible distortion, they leave distinct statistical artifacts that betray the removal attempt. We name this overlooked axis Watermark Removal Detection (WRD) and demonstrate that a modern classifier trained on these artifacts achieves state-of-the-art detection rates at $10^{-3}$ FPR