← All notes

AI search still needs the file permissions

· re: Foundry IQ SharePoint (indexed) knowledge source is now generally available (Microsoft)

Blueprint-style conceptual diagram of Microsoft’s SharePoint indexed retrieval. Source documents and supported permissions feed indexed chunks with source ACL metadata. A retrieval filter uses the requesting user and indexed permissions to return matching chunks to a permitted user and exclude restricted chunks for a user without access. Notes state that this is not the full SharePoint policy model and permission changes require synchronisation, with inherited changes sometimes needing resync.

AI search needs to respect the document’s permissions.

Microsoft’s 18 November announcement makes the SharePoint indexed knowledge source in Foundry IQ generally available. Supported user and group permissions are stored with indexed content, and retrieval filters the document chunks for the person asking.

Supported matters here. Microsoft says this does not reproduce the full SharePoint policy model. Permission changes appear after a successful sync, and inherited changes can need an explicit resync.

(Making a restricted folder easier to find was probably not the project brief.)

My read: the useful test for an AI Data Enclave is whether the document’s permissions also govern its search records. An AI acting for you should not retrieve records from a document you cannot read.

Put two users with different access through the same question. Check which source records each can retrieve before declaring the demonstration a success.