AIF-C01 Practice Question 68: Stale Knowledge Base Answers

Amazon Bedrock is explicit about this: *each time you add, modify, or remove files from your data source, you must sync the data source so that it is re-indexed to the knowledge base*. Ingestion is what converts raw files into vector embeddings, so until a sync runs (Sync in the console, or a StartIngestionJob request) the vector store still holds embeddings of the *old* text and retrieval faithfully returns it. Syncing is incremental, so only the changed documents are reprocessed (option C).

- A is wrong because there is no separate response cache to clear. What the assistant retrieves is whatever was embedded and indexed at the last ingestion; the staleness is in the index itself, not in front of it. - B is wrong because the training cutoff governs the model's own pre-trained knowledge. Sidestepping that limit is the whole point of RAG; the answer is supposed to come from retrieved passages, so the cutoff cannot explain stale runbook wording. - D is wrong because a guardrail blocks or masks content. The symptom would be a refusal or a redacted response, not a fluent, confident answer quoting the previous version of the text.

Reference: Sync your data with your Amazon Bedrock knowledge base