Canadian computer vision researchers presented a scientific paper on July 24, 2025, describing a new tool called "UnMarker" that can effectively remove protective watermarks from AI-generated images. This development, reported by Axios, poses a serious challenge to global efforts to ensure content origin transparency and combat the spread of disinformation and deepfakes. As leading AI companies like Google, Meta, and OpenAI implement visible and invisible watermarking technologies (e.g., based on the C2PA standard) for mandatory labeling of synthetic content, counter-technologies aimed at neutralizing them are also emerging. "UnMarker" is one of the most advanced in this line. Its key feature is that it does not need to know the specific watermarking algorithm used beforehand. The tool apparently uses its own diffusion AI model to find and intelligently "inpaint" anomalies in pixels or frequency characteristics of the image that are typical of watermarks, restoring the image to a "clean" state. This news is another turn in the continuous "arms race" between content authentication and forgery technologies. The existence of a universal tool like "UnMarker" significantly complicates the task of automatically identifying AI-generated images and proves that relying solely on watermarking technology may not be enough. This underscores the urgent need for a comprehensive, multi-layered approach to the problem, which includes legislative regulation, strict policies at the platform level, and improving overall media literacy in society.
Researchers Create "UnMarker" – An AI Tool to Remove Watermarks
