Definition
Permission-aware retrieval
Permission-aware retrieval is a retrieval architecture in which every candidate result is evaluated against the requesting user's real identity and access policies at query time, before any content reaches the language model. The model cannot leak what it never received.
The term names the missing property in most RAG systems: retrieval that knows who is asking. Vector search alone optimizes for similarity. IAM alone authorizes the application, not the person. Permission-aware retrieval joins the two, per query, per chunk.
Gateco evaluates each retrieved candidate against policy at request time and refills; it does not rewrite your vector query.
The late-binding architecture
Authorization binds at retrieval time, not at indexing time. That single choice is what makes revocation and policy changes re-index-free.
1. Permission-blind recall
The query runs against the vector database normally (vector, keyword, or hybrid search) and returns the top candidate chunks by relevance. Recall stays complete: nothing relevant is missed because of pre-filtering.
2. Per-chunk policy evaluation
Each candidate is evaluated against the requesting principal: their groups and attributes synced from the identity provider, the resource's classification and metadata, and the active policies (RBAC, ABAC, or relationship-based). The default is deny: a chunk with no matching allow policy is dropped.
3. Secure resolution and audit
Only the authorized subset is returned to the application or synthesized into an answer. Every decision is recorded: principal, resource, policy, verdict, and timestamp. Denials are visible as metadata, never as content.
Two users asking the identical question get different answers, each complete within their clearance. The pattern is described in depth in pre-retrieval vs post-retrieval authorization and demonstrated end to end in the recorded product demo.
Who implements late-binding authorization
Several products bind authorization at retrieval time. They differ in where the check runs and whose index it protects:
Microsoft 365 Copilot and Microsoft Search
Security trimming at query time with the signed-in user's token, inside Microsoft 365. Governance stays in SharePoint; retrieval goes through the Copilot Retrieval API.
Azure AI Search
Query-time ACL and RBAC enforcement (generally available with the 2026-10-01 REST API) and Purview sensitivity labels (preview), evaluated against permission metadata synchronized into the index; source-side changes appear after the next synchronization.
Fine-grained authorization engines (Auth0 FGA / OpenFGA, Cerbos)
A policy or relationship engine you call around your own retriever, as a per-document check or a query-plan filter. The retrieval wiring, the refill and the audit record are yours to build.
Enterprise search platforms (Glean)
Permission-aware search inside the platform's own index, for content the platform crawled.
Gateco
Vendor-neutral, in front of the vector stores you already run: evaluates each retrieved candidate against policy at request time, refills when candidates are denied, denies by default, and records every allow and deny. It does not rewrite your vector query, and it does not cover the surfaces listed below.
Comparisons with sources: Microsoft Purview and Azure AI Search, Cerbos, Glean.
What it deliberately does not solve
Permission-aware retrieval is one layer of an AI security program, not the program. Claiming otherwise would be wrong, so here is the boundary:
Prompt injection and jailbreaks
Permission-aware retrieval bounds what the model can see, which limits the blast radius of an injected prompt, but detecting and neutralizing adversarial instructions is a separate control.
Model security and red-teaming
Training-data extraction, fine-tuning integrity, and model supply chain are model-layer concerns, not retrieval-layer ones.
Output moderation and DLP on non-retrieval channels
If users paste sensitive data into prompts or the model generates from its own weights, retrieval enforcement never sees it.
Network and infrastructure authorization
Service-to-service authz, API gateways, and workload identity remain the domain of IAM and policy engines such as OPA.
Common questions
What is permission-aware retrieval?
Permission-aware retrieval is a retrieval architecture for AI systems in which every candidate result is evaluated against the requesting user's identity and access policies at query time, before any content reaches the language model. Content the user is not authorized to see is removed from the context, so the model cannot leak what it never received.
What is late-binding authorization in RAG?
Late-binding authorization defers the access decision to the moment of retrieval instead of baking permissions into the index. The pipeline runs a permission-blind similarity search, then evaluates each candidate chunk against the current identity, groups, attributes, and policies, then passes only the authorized subset onward. Because decisions bind late, policy changes and access revocations take effect on the next query once the change has reached Gateco, without re-indexing.
Who implements late-binding authorization for AI retrieval?
Microsoft 365 Copilot and Microsoft Search security-trim results at query time with the user's token inside Microsoft 365; Azure AI Search enforces ACLs, RBAC scopes and Purview labels at query time against permission metadata synchronized into its index; fine-grained authorization engines such as Auth0 FGA / OpenFGA and Cerbos are called around a retriever you build; enterprise search platforms such as Glean enforce inside their own index. Gateco is a vendor-neutral implementation in front of the vector stores you already run: it evaluates each retrieved candidate against policy at request time and refills; it does not rewrite your vector query.
How is permission-aware retrieval different from filtering the LLM's answer afterwards?
Post-generation filtering inspects the model's output after unauthorized content has already entered the context window, where it can be paraphrased, summarized, or blended beyond recognition. Permission-aware retrieval enforces before generation: denied chunks never reach the model, so there is nothing to redact and an audit log records exactly which chunks were allowed and denied, for whom, and under which policy.
Gateco implements permission-aware retrieval
Identity sync from your IDP, per-chunk policy evaluation, deny-by-default, and a full audit trail across 12 vector databases. When it is not the right fit, the RAG authorization guide says so honestly.