AI-WATERMARK

Summary from the TLDR InfoSec newsletter

UnMarker is designed to be the first universal attack against defensive watermarking, which is used to detect AI-generated deepfakes. The attack disrupts watermarks in image spectral amplitudes without requiring detector feedback or knowledge of the scheme, thereby defeating both traditional and advanced semantic watermarking systems and rendering deepfake detection ineffective.