Fact-checkers adopt cryptographic watermarks to detect AI deepfakes

Fact-checkers adopt cryptographic watermarks to detect AI deepfakes

Snopes debunked a viral hospital-bed image of Senator Mitch McConnell on July 8 by detecting Google's SynthID watermark embedded in the fabrication. The method marks a methodological shift for fact-checkers away from visual forensics toward cryptographic provenance checks. Google's SynthID and OpenAI's participating watermark system now offer verifiable detection, though the approach has gaps: Anthropic's non-participation leaves images from outside the watermarking consortium undetectable by this method alone.

Published

Read at another depth