The embedding of a signal into digital content so that its origin, integrity, or authorised use can later be verified, with the signal designed to survive routine processing while remaining imperceptible to ordinary consumers of the content.
In practiceWatermarks may be visible or invisible, robust or fragile, and may operate on images, audio, video, text, or model weights. In the generative AI context, providers use watermarking together with cryptographic content credentials to label machine-generated outputs, supporting downstream detection and disclosure obligations. No watermarking scheme is unbreakable: adversaries can paraphrase, recompress, or crop content, so watermarks are best understood as one layer in a wider provenance and disclosure regime rather than a standalone solution.
A media organisation configures its image-generation tool to embed an invisible watermark plus signed content credentials in every output, so that downstream platforms can flag the asset as synthetic even after the image has been resaved through a standard editor.
This definition is maintained by Moweb partners and used in live client engagements. For how Watermarking applies to your estate, or to challenge a working definition, speak to a partner.