Meta's Muse detector missed more than half of cropped images
Technology
Meta just launched an AI image detection tool for its Muse Image model, but it's not perfect, especially when images are heavily cropped.
The tool relies on an invisible watermark called Content Seal to spot Muse-generated images, but a Reuters analysis found it missed more than half of the cropped ones.
Experts say watermarking not foolproof
The watermark is meant to survive basic edits, but cropping can seriously weaken it.
Meta admits the tool works well on original images, but struggles when those images are trimmed down.
This highlights how tricky it is to keep up with manipulated AI content, and even Google and OpenAI face similar challenges with their own detection tools.
Experts say watermarking helps, but isn't a foolproof fix for deepfake worries.