Meta unveiled a preview of its new image‑generation model Muse Image alongside a detection tool that embeds an invisible watermark called Content Seal. The watermark is designed to survive common edits such as resizing or compression, providing proof that an image was created by Meta’s AI.

In a Reuters analysis of 40 AI‑generated images, the detector verified every original image but failed to confirm 55% after they were cropped to roughly one‑third to one‑half of their original size. The study highlights how simple edits can undermine watermark‑based verification.

Meta described the tool as a preview and acknowledged that the watermark may be compromised by heavy cropping. The company emphasized that its detection system is still evolving and is not yet foolproof.

Computer‑science professor Siwei Lyu noted that while watermarking can be highly effective, any modification that weakens the embedded signal—cropping, resizing, heavy compression—diminishes its reliability. AI researcher Sarah Barrington added that even catching 90% of cases is a significant improvement over zero, but it is not a complete solution.

Meta’s Oversight Board had earlier urged the company to address the “proliferation of deceptive AI‑generated content” and invest in stronger detection tools. With the U.S. midterms and other global elections on the horizon, the need for robust deepfake detection is more urgent than ever.

In sum, while Meta’s watermarking initiative marks progress, it also illustrates that a single technology cannot guarantee image authenticity. Multi‑layered detection strategies and industry‑wide standards will be essential to safeguard digital media integrity.