Meta's AI Detector Fails After Cropping | Spokane News

Meta’s new AI image detector, built to spot images from its Muse Image model using invisible watermarks, stumbled in real-world tests — missing 55% of cropped AI-generated images. While it caught original images flawlessly, simple edits like cropping significantly reduced its effectiveness, raising alarms ahead of key elections where deepfake detection is critical. Experts acknowledge watermark systems have limitations but still see them as major improvements over no detection at all. This comes after Meta’s Oversight Board had urged stronger safeguards — highlighting that despite recent moves, the tech still lags behind the speed of AI manipulation. Listen in comfort:Get a discount on a Soli Pillow: http://solipillow.com/discount/dnn. Advertise on DNN:[email protected] This is an automated, high-level news summary based on public reporting.Report issues to [email protected]. View sources & latest updates:https://sources.thednn.ai/281a479e28a6bef2