Meta’s preview AI image detection tool can fail to identify certain AI-generated images after they are cropped, according to a Reuters analysis cited by multiple outlets. The issue centers on Meta’s watermark-based approach: the detector performs well on the unedited images it generates with its Muse Image model, but its accuracy drops when the images are subjected to simple edits such as cropping. Several reports say researchers found that a large share of cropped versions of Meta-generated AI images are not flagged by the detection system. This limitation is presented as a technical constraint rather than a deliberate failure, and it highlights how common image modifications can interfere with watermark signals used for identification. The coverage also notes that Meta has been developing detection tools alongside its image generation features, in response to wider concerns about deepfakes and synthetic media. Overall, the sources agree that the tool’s effectiveness depends on preserving the watermark or detectable signal, and that straightforward edits can reduce or remove that signal enough to evade detection.