Knowledge Currency

Explanation

Knowledge currency checks that the documents and embeddings powering a RAG system are fresh, so the assistant answers from up-to-date sources rather than stale ones — for example a banking policy chatbot must not quote a withdrawn interest-rate circular because the vector store was never refreshed. It matters because stale knowledge produces wrong, potentially non-compliant answers, and the control carries the proposed RBI Safety/Resilience, NIST MANAGE 3.1 and ISO Clause 6.1.2 (privacy) mappings (no fixed priority is set). It is tracked through a Vector Freshness SLA; no calculation method is specified in the control, so operationalise it by timestamping each source document and its embedding and continuously measuring the age of indexed content against the agreed SLA window. To implement, instrument the RAG ingestion and re-indexing pipeline to record when each item was last updated versus last re-embedded, expose the oldest/stale-item age as a monitored metric, and log refresh runs and any items exceeding the window as evidence. The threshold is that vector freshness stays within a defined staleness window — for example 24-72 hours — with an alert on breach (this is proposed). When indexed content ages beyond the window, the breach triggers an alert and a re-indexing / refresh job, and stale content should be flagged or held back until updated so answers are not built on out-of-date knowledge.