UnMarker is a novel universal attack on defensive image watermarking that operates without needing detector feedback or advanced knowledge of the watermarking schemes. It employs two unique adversarial optimizations to effectively erase watermarks from images, demonstrating significant success against various state-of-the-art watermarking methods, including semantic watermarks that are crucial for deepfake detection. The findings challenge the efficacy of defensive watermarking as a viable solution against deepfakes, highlighting the need for alternative approaches.