Meta’s AI Detector Misses 55% of Cropped Images, Exposing Watermark Weakness
In Reuters tests, Meta's Content Seal watermarking correctly identified all 40 original AI images from its Muse Image model but failed on 55% after moderate cropping. The results underscore the fragility of embedded detection methods and the urgent need for robust content authentication in an election year.
Source: britainnews.net