Provenance in the AI pipeline
Provenance is the documented history of a creative work. It answers the questions: where did this come from, who created it, how has it been used, and on what terms? In the age of AI, it has become an operational necessity.
The provenance gap in AI
Most AI systems currently operate with minimal provenance information about their training data. A model trained on a web scrape may have processed millions of creative works with no record of who created them or whether consent was granted.
The provenance chain
1. Creation — the CDR is established, capturing rights, consent, and input licence class
2. Ingestion — the platform reads the CDR, checks consent, and records the ingestion event
3. Transformation — each transformation is recorded against the relevant class
4. Output — the output carries a Provenance Certificate linking back to originating CDR(s)
5. Distribution — the Provenance Certificate travels with the output
What good provenance infrastructure looks like
- For creators — an active CDR in the Rights Registry, with cip.md declaring your rights
- For platforms — 95% Rights Payload coverage, audit logging, and Provenance Certificates on all outputs
- For agencies — portfolio-level CDR maintenance across all client assets
- For lawyers — contract clauses that require provenance documentation as a condition of licensing