A company wants to retrieve patient history details to augment the Al agent response. [he company wants to leverage the Data Cloud search index feature. What is best practice when considering retrieva

The correct answer is C. Mask sensitive fields and index only non-PII data.. When using Retrieval-Augmented Generation (RAG) with Data Cloud's search index, best practice for PII is to avoid indexing it altogether. Sensitive fields (e.g., SSNs, dates of birth, contact details) should be masked or excluded before data is indexed, so they are never embedded

AI Features for Service (e.g., Einstein Bots, Next Best Action)

Question

A company wants to retrieve patient history details to augment the Al agent response. [he company wants to leverage the Data Cloud search index feature. What is best practice when considering retrieval-augrmented generation (RAG) for information that may contain personally identifiable information (PII)?

Options

ADepend on the agent's prompt to avoid exposing PII.
BEncrypt embeddings, but still index PII records.
CMask sensitive fields and index only non-PII data.

How the community answered

(55 responses)

A
9% (5)
B
18% (10)
C
73% (40)

Explanation

When using Retrieval-Augmented Generation (RAG) with Data Cloud's search index, best practice for PII is to avoid indexing it altogether. Sensitive fields (e.g., SSNs, dates of birth, contact details) should be masked or excluded before data is indexed, so they are never embedded, stored in vector form, or surfaced in agent responses. Option A is insufficient because prompt instructions are not a security control and can be bypassed or ignored. Option B is worse - encrypting embeddings while still indexing PII means the PII data is still present in the index and could be retrieved; encryption at the embedding level does not prevent the content from appearing in LLM responses.

Topics

#RAG#PII#Data Security#AI Best Practices

Community Discussion

No community discussion yet for this question.

Full AI-201 Practice