A Reuters analysis found that Meta’s AI image detection tool, introduced alongside its Muse Image model, failed to identify 55 percent of AI-generated images after they were cropped to about one-third to one-half of their original size. Of 40 images tested, the detector correctly verified all of the original, uncropped versions but missed most of the cropped ones. The tool relies on an invisible watermark called Content Seal, which Meta acknowledges can be degraded or lost through common edits such as cropping. The finding highlights limitations in watermark-based AI detection ahead of a heavy election cycle that includes the US midterms.
