Crew incident reports at Meridian Line are filed as free text and often include a crew member's name, passport number, or home address alongside the safety details. Before this text is extracted, chunked, and landed in the Delta table that feeds the safety-review summarizer's retrieval index, what should happen to those fields?
Select an answer to reveal the explanation.
Short Explanation
Fix it at the source, not at the exit door. If PII stays out of the table the index is built from, there's no query, no prompt, and no summarizer response that can ever surface it — you're not relying on catching it every single time on the way out.
Full Explanation
Masking or redacting PII during data preparation — before the crew-report text is written to the Delta table that becomes the retrieval corpus — means the sensitive fields are never present in the index at all, so no query, retrieval, or generated summary can surface them; this is a structural fix rather than something that depends on catching every leak downstream. Access control on the Vector Search index restricts who can query it, but it does nothing about what's inside the indexed content itself — someone with legitimate access to the safety-review tool still shouldn't be shown a crew member's passport number embedded in a retrieved chunk, and access control alone doesn't stop the PII from being embedded and retrievable in the first place. Relying on stripping PII only at final-response time, on every single query, means the PII sits in the Delta table and gets embedded into the vector index regardless, and depends on a redaction step never being missed on any of the many paths a chain or agent might expose retrieved content — a single unguarded path (a raw-context debug view, a differently-formatted tool output) leaks it. Reversible encryption keeps the underlying PII fully recoverable by definition, which doesn't satisfy the intent of masking it out of a document corpus that's meant to be searched and summarized as text — encrypted text also isn't usable for retrieval or embedding without decrypting it, defeating the purpose. Exam framing: this is the Governance objective 'apply data-masking and PII-protection techniques to source data and to application inputs/outputs,' and it deliberately ties back to the data-preparation stage rather than treating masking as a serving-time afterthought.