Meta's New AI Image Detector Fails on Cropped Images, Reuters Finds

A new artificial intelligence detection tool from Meta Platforms failed to identify a majority of its own AI-generated images after they were cropped, a Reuters analysis found. The results underscore the significant technical hurdles in verifying manipulated digital content, a critical issue for investors and the public amid a rise in sophisticated deepfakes during a U.S. election year.

Analysis Reveals Watermark Weakness

The tool, which Meta previewed this week alongside its new image-generation model, Muse Image, is designed to identify AI-created pictures using an invisible watermark called Content Seal. Reuters tested 40 images made with Muse Image and found the detector successfully verified all the originals.

However, after the images were cropped to approximately one-third to one-half of their original size, the tool failed to identify 55% of them as AI-generated. In response to the findings, Meta noted that the tool is a preview and that while the watermark is designed to withstand common edits, its signal can be lost if an image is heavily cropped.

A Persistent Industry Challenge

The difficulty in reliably detecting altered AI content is not unique to Meta. Competing technology firms like Google and OpenAI have also cautioned that their own detection systems are not foolproof against common image modifications. The inability to consistently identify synthetic media poses risks to information integrity, which can impact public discourse and market stability.

Ad

Experts in the field confirm the inherent limitations of current technology. Siwei Lyu, a computer science professor at the State University of New York at Buffalo, told Reuters that watermark-based systems are effective only when the watermark remains intact. Modifications like cropping, resizing, or heavy compression can weaken or remove the embedded signal.

Context and Regulatory Scrutiny

The findings come as Meta faces pressure to enhance its content moderation capabilities. In March, the company's own Oversight Board urged it to invest in stronger detection tools to address the "proliferation of deceptive AI-generated content" on its social media platforms.

While watermarking technology holds promise, researchers acknowledge its current state. "It may not be fully watertight, but even if we catch only 90% of cases, that’s still a great leap from 0,” said Sarah Barrington, an AI researcher at the UC Berkeley School of Information, in comments to Reuters.