Reuters reports that Meta’s AI image detector works inconsistently when images are modified by cropping. In an analysis of 40 images produced using Meta’s Muse Image, the detector successfully identified all of the original AI-generated images. However, when the same images are cropped, the tool fails to verify a large share of them. The analysis finds that the detector does not identify 55% of the cropped versions, even though they come from the same source images initially generated by Muse Image. The findings indicate that the detection performance drops substantially after simple image edits rather than after changes like re-encoding or applying new filters. The reporting also highlights that the test focuses on cropped versions of AI-generated images, comparing verification results before and after the edit. Overall, the issue centers on detection accuracy degrading for cropped AI images generated by Meta’s own system, suggesting limitations in how robust the detector is to basic alterations.