Your access control is probably correct and your model is probably still leaking. That is the uncomfortable part of llm knowledge leakage prevention: the permission check runs at the document store, a

Markdown supported · ⌘↵ to post
No responses yet.