Last updated:
The practice of embedding detectable, machine-readable provenance markers in AI-generated output, such as images, audio, video, and synthetic text, so downstream consumers and platforms can identify content as AI-generated.
Watermarking is the technical foundation for AI content provenance and a regulatory obligation under the EU AI Act's Digital Omnibus, taking effect 2 December 2026. Watermarks must survive common transformations such as copy-paste, screenshots, and reformatting to provide reliable provenance signals. The detailed technical standard for AI Act watermarking is expected via implementing acts; the C2PA Content Credentials standard is the likely reference point. Other major jurisdictions including China and the United Kingdom have introduced parallel obligations. Engineering teams shipping generative AI output should treat watermarking as a near-term capability requirement rather than a future option.
Real world example:
A SaaS marketing platform generates email copy and social images using a foundation model. Under the AI Act, the platform must embed watermarks in every AI-generated output it returns to customers from 2 December 2026. The engineering team scopes a watermarking proof-of-concept in summer 2026, prioritizing image and audio output first because these have established provenance standards (C2PA) and the clearest regulatory expectations.
