Advanced security engineering

Follow the trust.
Test the boundary.

Work through the difficult cases: a token crosses services, a revoked document survives in a cache, a tool changes after approval, or two events appear related.

18 deep-dive guides29 advanced hunts3 connected paths
Choose an investigation →
Event time, collection time and matching identity are separate dimensions of an investigation.
Build a chain of evidence. Record where the inference stops.

Start with the hard question

Three routes into the deeper work.

Identity & delegation

Who is really authorized?

Trace user delegation and application authority separately. Investigate federation trust, consent and credential changes.

Work toward: An authority map, token-validation cases and a scoped investigation.

Follow the path →

Retrieval & ingestion

What enters the context?

Follow permissions through document ingestion, chunking, vector retrieval, caches and revocation.

Work toward: A document-boundary test matrix and a verified revocation plan.

Follow the path →

Tools & adversarial evaluation

What can an agent cause?

Bind approvals to real actions. Check MCP tool authority, gateway policies and evaluation evidence.

Work toward: An action-boundary design and repeatable tests of actual side effects.

Follow the path →

Architecture, failure modes & validation

Deep-dive guides.

Filter the guide library →
Retrieval & data security5 min read
Advanced guide

Secure the storage data plane behind AI retrieval

Separate ingestion and serving authority, close Shared Key and anonymous paths, constrain SAS use, and validate Blob and Data Lake access independently.

Documentation checked
Identity & authorization5 min read
Advanced guide

Design Conditional Access around AI identity types

Separate human, service-principal, managed-identity, and preview agent policy boundaries; validate token acquisition and resource enforcement independently.

Documentation checked
Identity & authorization7 min read
Advanced guide

Design delegated authorization across an agent tool chain

Separate user delegation from application authority, validate token boundaries, and prevent an agent from turning an authorized session into an unauthorized action.

Documentation checked
Agents7 min read
Advanced guide

Engineer MCP tool trust boundaries beyond the prompt

Design per-request authorization, tool admission, downstream identity and consequential-action approval for remote MCP integrations with Microsoft Foundry.

Documentation checked
Identity & authorization5 min read
Advanced guide

Constrain privileged access across the AI control plane

Design PIM activation, resource scope, approval evidence, and emergency recovery around the permissions that can change models, data, identities, and telemetry.

Documentation checked
Retrieval & data security7 min read
Advanced guide

Enforce document authorization before RAG grounding

Choose between application security filters and native preview permissions, preserve access metadata on chunks, and enforce the same boundary across keyword and vector retrieval.

Documentation checked
Retrieval & data security7 min read
Advanced guide

Protect RAG ingestion against poisoned and untraceable content

Separate source access from content trust, preserve provenance through chunking, and verify publication, deletion, and rollback without assuming a prompt filter proves integrity.

Documentation checked
Detection & response5 min read
Advanced guide

Validate Sentinel detections beyond a successful KQL parse

Test schema assumptions, event semantics, ingestion timing, join behavior, scheduling, entity mapping, and analyst outcomes before promoting a hunt into a detection.

Documentation checked

Each guide separates product behavior, architectural recommendations and test evidence. Check its prerequisites and preview limitations before applying an example.

From a signal to a defensible finding

Advanced hunting recipes.

Filter the query library →
Static checks are one part of validation.

These queries are documentation-checked hunting starters. Validate data coverage, stable identity keys, time windows, expected results and false positives in your own workspace before operational use.

Read the correlation engineering guide →