How do you leave an audit trail without copying confidential data into every log?
Review note: The answer refers to the original author's CV or target job offer — adapt it to your own experience.
Answer out loud or in your head first, then compare.
Answer
I record task identifiers, versions, access decisions and the operation's outcome. I keep the full content only when it's needed and properly protected. Not everyone looking at metrics should be able to read employees' documents.
For an approval I need to know who approved which proposal. For an error, the document and version identifiers are often enough, which an authorised person can reconstruct in a controlled place. I set retention and deletion of data. Automatic masking of secrets helps, but it isn't flawless. Product analytics, diagnostics and auditing may need different scopes of access.
Why it matters: Confidentiality also applies to the tools for observing the system.
Possible follow-ups: Who can reconstruct the model's full input, and when should they be able to?
How did it go? Be honest — “Not yet” puts the question on your review list.
Start with
- J-P01 · What is evaluating an AI system, i.e. an eval? — Evaluation, quality and production AI
- J-P02 · How does an eval differ from an ordinary code test? — Evaluation, quality and production AI
Further reading
- Azure AI Search: Document-level access control(opens in a new tab)Docs; accessed 22.10.2026. Security filters, native ACLs and preview features; syncing permissions needs an explicit mechanism.