Meta launched Content Seal on July 7, an invisible watermarking system for images and videos created by its Muse AI models, designed to survive cropping, resizing, and compression. However, independent testing by Reuters on July 10 found that Content Seal failed to identify over 55% of cropped AI-generated images. The system correctly identified original, unmodified images but missed more than half of cropped versions. Meta also launched a web-based detection tool at meta.ai/identification. The rollout followed pressure from Meta’s Oversight Board in March to combat deceptive generative AI content. Content Seal operates outside existing industry standards, including the C2PA coalition’s Content Credentials and Google’s SynthID, creating fragmentation for platforms and fact-checkers.