WHITEBOXX / GLOSSARY

AI provenance: the history
of an AI result.

AI provenance is the record of how an AI result was produced: which inputs it used, which model and tools produced it, who requested or approved it, and when. It lets someone check where a result came from without having to trust the system that made it.

By Gautam Mehra · Published · Updated

What is AI provenance?

The term is used in two ways. For AI-generated media, provenance means labelling content so people can tell it was made or edited by AI. For AI used in business work, it means keeping the history of each result so it can be traced and checked.

Both rest on the same idea as provenance in art or science: a documented chain from origin to final form. The W3C published a general standard for describing it, PROV, in 2013.

STANDARDS

Provenance standards and rules.

STANDARDS
StandardWhat it covers
W3C PROV (2013)A general model for describing the things, activities and people involved in producing something
C2PA Content CredentialsSigned metadata attached to images, video and documents that records how they were made or edited
EU AI Act, Article 50Requires providers to mark AI-generated content in a machine-readable way, with some exceptions

Why does AI provenance matter?

Generated text and figures look equally plausible whether they are right or wrong. Provenance turns “the system said so” into a chain someone can check: this figure came from that source, retrieved on that date, through this calculation.

For enterprises, provenance also settles accountability. It shows which person asked for a result and which person approved it.

How Evidence OS records provenance.

Evidence OS results carry the sources and calculations behind them, and each run records its inputs, steps and deliverables. A reviewer can trace a finding back through the run that produced it.

See the audit trail