Fact-checkers turn to cryptographic watermarks to identify AI deepfakes

Snopes used Google's SynthID watermark to debunk a fabricated image of Senator McConnell this month, signaling a shift in fact-checking methodology. Rather than analyzing pixels for signs of manipulation, checkers now verify cryptographic signatures embedded by AI generators. Google and OpenAI's systems enable reliable detection, though Anthropic's non-participation creates blind spots for images outside the watermarking ecosystem.