AI Watermarking in plain English.

AI watermarking embeds an invisible signal into AI-generated images, video, audio, or text specifically so the content can later be identified as machine-made, even after some editing. Google's SynthID is the most widely deployed example, built into Gemini's image and video outputs; the C2PA standard (backed by Adobe, Microsoft, OpenAI, and others) takes a related but different approach, attaching cryptographically signed metadata about an image's origin and edit history rather than altering the pixels themselves. Watermarking is meant to help platforms and users tell real footage from synthetic footage, particularly around elections and breaking news, but it has real limits: screenshots, format conversions, and deliberate stripping can destroy watermarks, and there's no equivalent industry standard yet for plain text, which is far easier to paraphrase past detection. Watermarking is best understood as one layer of provenance, not a guaranteed way to catch AI content.