AI Watermarking
Working knowledgeGovernance, Safety and Ethics
Also called: Watermarking (AI Output), Watermarking
AI watermarking is the practice of embedding subtle, mathematically detectable signals or metadata into text, images, audio or video generated by a model. The mark is invisible to a reader but recoverable by a checking tool, so an output can later be identified as synthetic and traced back to its origin.
In practice
Watermarking is how a business verifies media authenticity, detects deepfakes and meets disclosure rules that require artificially generated content to be labelled. Two questions decide whether it is worth anything to you: does the tool generating your content mark its output, and does anyone check? A mark nobody looks for protects nothing.