Meta AI Image Detector Fails to Verify Its Own Images

The Meta AI detection tool failed to identify many cropped AI-generated images during a Reuters analysis. The findings raise questions about AI image verification and deepfake detection. The issue comes as online platforms prepare for a major election period that includes the U.S. midterms.

Reuters tested 40 images created with Meta's Muse Image model. The Meta AI detection tool successfully verified every original image. However, it failed to verify 55% of the same images after researchers cropped them. The cropped versions measured about one-third to one-half of their original size.

Meta says the preview system uses an invisible watermark called Content Seal. The company embeds this watermark into every image produced by Muse Image. According to Meta, the system should verify images even after common edits, including cropping.

After Reuters shared its findings, Meta emphasized that the Meta AI detection tool remains a preview. The company added that heavy cropping may weaken or remove the watermark signal. That limitation can reduce successful verification.

Google and OpenAI have also warned that their AI detection systems cannot reliably resist image alteration methods. Those companies acknowledge that modified images remain difficult to verify consistently.

In March, Meta's Oversight Board urged the company to strengthen protections against deceptive AI-generated content. The board recommended further investment in detection technology across Meta's platforms.

Siwei Lyu, a computer science professor at the State University of New York at Buffalo, said watermark systems have limits. He explained that cropping, resizing, compression, and editing can reduce watermark performance, depending on the design.

Also Read: Meta Rolls Out Muse Image Across Meta AI Apps

Sarah Barrington, an AI researcher and Ph.D. candidate at the UC Berkeley School of Information, said watermarking remains promising. She added that detecting even 90% of cases would represent a significant improvement over detecting none. The Meta AI detection tool highlights both the progress and remaining challenges in AI image verification.