Research · Data for AI

Data Governance & Lineage for AI

Governance defines who owns data, what it means and how it may be used. Lineage traces where data came from and how it changed. For AI, both are needed to explain outputs, meet obligations and fix problems.

What changes with AI

• Data is reused for purposes it was not originally collected for, such as model training.
• Unstructured content becomes a governed asset, not just files.
• Outputs must be traceable to training data, retrieved sources and model versions.
• Regulations increasingly require documentation of data used in AI systems.

Core elements

• Named owners for data domains and knowledge sources used by AI.
• A catalog showing what data exists, its sensitivity and approved uses.
• Lineage from source systems through preparation to models and outputs.
• Records of which data and model versions produced a given result.
• Policies on retention and deletion that reach AI training and retrieval stores.

Why it matters

Without lineage, teams cannot answer basic questions after an incident: which data caused a biased result, whether a deleted record is still in a model, or which systems used a data set that turned out to be wrong. Governance also speeds approval, because reviewers can see what data a new AI use would touch.

Common pitfalls

• Governance programs that cover databases but not documents.
• No record of which data trained or grounds each model.
• Deletion requests that never reach AI data stores.

How to get started

• Extend your data catalog to cover AI training and retrieval sources.
• Record data and model versions for every production AI system.
• Define approved AI uses for sensitive data classes.
• Test that deletion and retention rules reach AI systems.