Provenance for human and AI created work

As more digital content is created with generative AI, questions about authorship, copyright, intellectual property and governance are getting harder to answer.

Those questions depend on reliable information about who was involved, which tools were used, and how the work was created. In practice that information is often incomplete or missing.

A workflow-level approach

ProvenanceKit grew out of research by Baranaba Mugabane in the AI for Sustainable Societies master's programme, which designed and evaluated a governance-oriented provenance framework for human and AI created works.

The research introduces a workflow-level approach built on two ideas:

• The Entity, Action, Attribution model. Every record is a small graph of entities (people, AI models, organisations), the actions they took (create, edit, remix, verify) and the attributions that link contributions to outputs.
• An extensible metadata structure. Typed extensions capture what matters for governance: AI involvement, rights, licences and authorisation conditions.

The design combines legal analysis across the EU, US and WIPO frameworks with the technical work of recording provenance as content is made, rather than reconstructing it afterwards.

Why it matters

Transparency is a precondition for fairness. When the chain of creation is recorded, credit can follow contribution, rights can be respected, and people can decide how far to trust what they see.

ProvenanceKit is open source. You can read more at provenancekit.com and start building with the documentation.