Closing an Azure OpenAI assistant's retrieval gap didn't take a new identity platform. It took one filter and a narrower assistant.

SynSphere Italia CEO Egiziago Cioffi found that his Azure OpenAI email assistant, despite passing evaluations and resolving about 60% of inbound customer emails, returned SharePoint content to users who lacked access. The issue reflected a common RAG failure mode: the assistant retrieved data using the indexer’s permissions rather than the requester’s, and some Azure access-control paths remain incomplete or fail open if not configured correctly.

The case shows that AI agent security depends less on broad identity overhauls and more on enforcing authorization at retrieval time. As enterprises deploy copilots over internal data, permission-aware retrieval and narrower assistant design are becoming baseline requirements rather than optional hardening.