Chapter 1 of 6
Session scope#
What we will do#
Objective. Record approved models, deploy their exact serverless API versions, and apply the matching built-in deployment policies. Version-controlled profiles and the approval register hold the inputs. Preflight compares them with current Azure state and scoped Bicep previews.
The observable result is a live deployment whose model coordinates, SKU, capacity, and approval tag match the approved profile.
Why it matters#
Problem. An unreviewed model version or deployment type can move data to the wrong processing location, burn quota, or expose the workload to retirement without warning.
Solution. This session checks those choices against live Azure state before deployment. It then uses the approval register to assign Azure Policy controls for approved models and model eligibility.
Boundaries#
Azure is authoritative for availability, quota, lifecycle data, and live deployment state. The customer decision system holds the full approval and review history. This repository owns the deployment profile and Bicep definition.
The deployment profile governs this Bicep path. The built-in Azure Policy assignments evaluate model deployments at the approved resource group. The policy owner reviews Audit findings before moving the assignments to Deny.
This session excludes instant-access and managed-compute models. The platform baseline and private-networking controls establish the Foundry baseline and private path. The evaluation gate controls release eligibility.
Session preparation
Who should join
- AI product owners and AI engineers
- Cloud engineers who run Microsoft Foundry
- Data, privacy, procurement, and security reviewers
- Azure Policy owners
What you need
- An
AIServicesMicrosoft Foundry resource and project exist in the approved nonproduction subscription and resource group. The platform owner confirms the resource and project names, and that the approved execution host reaches the project through the recorded private path. (Sessions 01 and 02.) - Record the existing Microsoft Foundry resource and project names, and confirm that the resource
kindisAIServices. - Give the deployment operator time-bound Cognitive Services Contributor on the exact Microsoft Foundry resource.
- Give the policy operator time-bound Owner or Resource Policy Contributor on the approved resource group.
- The model decision authority approves the exact model coordinates, workload purpose, processing-location requirement, and external decision reference through the customer model-change process.
- Confirm that the platform owner can read model availability and quota, and that the named Responsible AI policy exists under the exact Foundry resource.