Meta’s AI image detector fails to catch 55% of its own cropped AI images

Meta’s new AI detection tool failed to identify more than half of cropped images generated by its own image model, according to a Reuters analysis.

The tool was previewed this week alongside Muse Image, Meta’s new image generation model. Reuters tested 40 images made with Muse Image and found that the detector verified all of the original outputs. But once the same images were cropped to about one third to one half of their original size, the tool failed to verify 55% of them.

Meta says the detector relies on Content Seal, an invisible watermarking system embedded in images created by Muse Image. The company says the watermark is designed to help users verify whether an image was created by Meta’s AI models, including after common edits such as cropping.

Meta told Reuters the tool is still in preview and said the watermark is meant to remain intact after common edits. The company said the signal may be lost if an image is heavily cropped.

The result points to a practical weakness in watermark based AI detection. Images shared online are often cropped, resized, compressed or edited before they circulate across social platforms.

Google and OpenAI have also warned that their own detection tools are not foolproof against image alteration techniques.

Meta’s Oversight Board called on the company in March to invest in stronger detection tools and do more to address deceptive AI generated content across its platforms.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.