AI or Not Detects 100 Percent of Meta AI Images in Benchmark Test

In a press release, AI or Not reported benchmark results showing that its detection technology identified 100 percent of Meta AI images in their original form and maintained 98 percent accuracy after cropping or tampering.

The benchmark tested 205 Meta AI images under two conditions: original and cropped or tampered. Each image was analyzed by both Meta AI's labeling system and AI or Not's detection API. On original images, Meta AI's labeling correctly identified 98.1 percent, while AI or Not achieved 100 percent. After cropping and tampering edits were applied, AI or Not's accuracy remained at 98 percent, compared to Meta AI's 35.3 percent.

According to the company, the test demonstrates that detection technology trained against adversarial manipulation remains reliable even when watermarks or metadata are removed.