Meta's new tool failed upon launch? It can't even recognize the images you created.
- bySherya
- 13 Jul, 2026
Meta AI Detection Tool: Meta recently previewed an AI detection tool, and it failed in initial trials. The tool failed to recognize cropped AI-generated images.

Meta AI Detection Tool failed to recognize its own image.
Meta AI Detection Tool: Meta's new AI detection tool failed to recognize its own images. Last week, Meta teased a new AI detection tool with its new image generation model, Muse Image . Now, it's been revealed that the tool can't recognize its own images. Reportedly, cropping an AI-generated image makes it difficult for the tool to recognize it.
Trouble recognizing cropped images
News agency Reuters tested this tool with 40 images. Images created using Muse Image were used in the trial. The tool recognized all 40 AI-generated images, but when presented with cropped images, it failed to detect 55 percent of them. This means its accuracy drops by more than half for cropped images. Meta, on the other hand, claims on its website that the model can also recognize cropped AI-generated images. According to Meta, the tool uses a watermarking system called Content Seal. Every image created using Muse Image has an invisible watermark. This system was used to detect AI-generated images.
Meta said this when the tool failed.
When Meta was asked about the tool's failure, the company stated that it was only a preview. The watermark is designed to remain even after cropping the photo, but it's possible that the signal was missed due to excessive cropping. It's worth noting that doubts always linger about the accuracy of AI detection tools. Companies like OpenAI and Google have warned that their AI tools aren't foolproof, meaning they can make mistakes in identifying AI-generated content.
Watermarks cannot always be trusted.
Experts say watermark-based detection systems have their limitations. Siwei Lyu, a computer science professor at the State University of New York, said that while watermarks are useful if they are intact with the photo, modifications like cropping, resizing, compression, and editing can cause the watermark's signal to disappear or become weak.






