Maintaining strong governance in rapidly evolving cloud environments is an ongoing challenge. As Azure services advance, policy frameworks must keep pace to safeguard security, compliance, and operational consistency without stifling innovation. The AzPolicyFactory project, led by Tao Yang, exemplifies how continuously refining Azure Policy definitions and initiatives empowers organizations to enforce nuanced controls across new and existing cloud workloads.

The recent AzPolicyFactory updates introduce a significant expansion in policy coverage, specifically targeting Microsoft Defender for Cloud (MDC), AI Landing Zone components, and Azure Kubernetes Service (AKS). These additions reflect the growing importance of securing AI services, adapting to Kubernetes' operational subtleties, and modernizing cloud workload protection. This post dives into the key policy improvements, how they integrate into existing governance frameworks, and practical insights for implementing them effectively.

Architecture Overview

┌────────────────────────────────────────────┐
│Architecture                                │
├────────────────────────────────────────────┤
│• Enterprise data sources                   │
│• Foundry platform                          │
│• AI applications                           │
└────────────────────────────────────────────┘

Key Technical Observations

• Up-to-date MDC Policies Using Current API Versions: The new Microsoft Defender for Cloud policies replace outdated built-in policies and leverage updated API versions, improving coverage and integration. This ensures better alignment with the evolving Defender features and Azure platform changes.

• Granular AI Foundry Controls Tailored to Emerging LLM Models: Policies restrict AI Foundry cognitive services to specific permitted large language model (LLM) formats (e.g., OpenAI, xAI, Anthropic) and approved model names (e.g., gpt-5, claude-opus-4-8), allowing organizations to tightly control AI workloads to mitigate risk and comply with governance.

• Separation of AKS Control Plane and Data Plane Policies: By splitting AKS policies into control plane and data plane initiatives, AzPolicyFactory optimizes policy testing performance and simplifies maintenance, recognizing the distinct operational scopes within Kubernetes clusters.

• Dynamic Policy Parameterization for ARM Preview API Restrictions: The updated policy restricting usage of -preview ARM API versions uses string parameters and request context conditions to enable exemptions by resource type and avoid false compliance negatives during scans, showcasing an advanced policy design pattern.

• Allow and Deny Lists for Resource Providers Enable Flexible Controls: Roles can start conservatively with allow lists when new to Azure or apply deny lists when mature, supporting progressive governance without disrupting existing workloads.

How It Works: Deep Dive into Key Updates

New Microsoft Defender for Cloud (MDC) Policies

AzPolicyFactory identified that the built-in Microsoft Defender for Cloud policies were outdated and replaced them with definitions based on the latest Azure API versions. These updated policies are grouped into an initiative polset-mdc.json that covers plan coverage and configuration settings more comprehensively.

This modernization provides enterprises with better compliance enforcement corresponding to current threats and Defender capabilities, ensuring cloud environments receive effective protection.

AI Landing Zone Policies for Emerging Cognitive Services

The AI Foundry set introduces several innovations:

• Network ACL Enforcement: Unlike the built-in policy, the new version mandates that networkAcls.defaultAction is set to Deny for restricting public network access, providing a stronger security posture for AI accounts.

• Authentication Controls: Instead of the built-in policy’s Modify effect, the new policy uses the Deny effect to outright prevent local authentication, reducing attack surface by enforcing stricter authentication mechanisms.

• Model Format and Name Restrictions: These policies maintain an allowlist of acceptable model formats and names, integrating vendor-specific and version-specific constraints to prevent unapproved AI model deployments. The initiative uses multiple instances of these policies, making it customizable per organizational needs.

Cosmos DB Policies Addressing Real-World Scenarios

One challenge with built-in Cosmos DB policies was handling absent capabilities. The updated policies include logic to handle empty arrays and block deployments when capabilities are undefined, avoiding gaps in enforcement.

Additional policies enforce minimum TLS versions and restrict key-based metadata write access with audit or deny effects, surpassing built-in append effects and providing parameterized control for administrators.

AKS Policies: Control and Data Plane Separation

Policies for managing AKS resources are divided into two coherent sets:

• Control Plane Policies (polset-aks-control.json): Cover API surface and management plane controls.

• Data Plane Policies (polset-aks-data.json): Contain built-in policies specifically targeting Microsoft.ContainerService.Data resources.

This separation streamlines policy management and testing by targeting distinct operational layers within AKS environments.

Advanced ARM Preview API Version Restriction

To address challenges in restricting usage of preview ARM REST API versions, the updated policy:

{
  "value": "[tryGet(requestContext().identity, 'idtyp')]",
  "in": ["user", "app"]
}

ensures the policy triggers only on operations initiated by users or applications, not during compliance scans where the identity is null. This guard prevents perpetual flagging of resources as non-compliant when the latest API versions happen to be preview.

The policy uses string parameters to target specific resource types, allowing incremental application and tailored exemptions.

Quick Tips & Tricks

1. Customize AI Model Policies to Match Organizational Risk Appetite
   Tailor the allowlist of model formats and names in the AI Foundry policies to tightly control which LLM services are available, avoiding accidental use of unvetted AI models.

2. Separate Policy Initiatives by Resource Plane for Performance
   When working with large resource sets like AKS, partition policies into control and data plane sets to optimize compliance scanning and testing speed.

3. Use Exemptions in Preview API Restriction Policy
   Leverage the string parameter and exemption capability in the updated ARM preview version policy to incrementally apply policy without impacting legacy workloads.

4. Start with Allowed Resource Provider Policies for New Azure Environments
   Use allow lists to whitelist essential resource providers early on, then gradually expand as the cloud estate matures for controlled growth.

5. Audit Before Denying with Cosmos DB Policies
   Initiate new Cosmos DB policies with the audit effect to understand current exposures before enforcing deny, ensuring smooth policy adoption.

6. Leverage AzPolicyLens for Automated Insights
   Combine AzPolicyFactory with AzPolicyLens to automate Azure Policy insights and reporting, simplifying governance visibility.

Conclusion

AzPolicyFactory’s latest updates deliver precise, modernized policy definitions and initiatives that reflect Azure's evolving service landscape. Enhanced AI Foundry policies position organizations to assert fine-grained control over rapidly developing cognitive services, while revamped MDC and AKS policies ensure governance keeps pace with security demands and operational complexity. The thoughtful improvements in ARM preview API version restriction policies highlight an advanced understanding of policy lifecycle challenges in real-world cloud environments.

Adopting these updated policies provides both tighter security controls and improved management flexibility, empowering organizations to confidently embrace Azure’s latest technologies without compromising compliance or operational excellence. Looking ahead, continuous evolution of policy toolsets like AzPolicyFactory will remain critical as cloud-native capabilities expand and regulatory requirements tighten.

References

1. Various updates for AzPolicyFactory - Tao Yang Blog — Original source covering all recent updates.
2. Policy Definitions for Microsoft Defender for Cloud — Up-to-date policy definitions for MDC.
3. AI Foundry Cognitive Service Policies Initiative — Policy initiative for AI Foundry cognitive services.
4. AKS Control Plane Policies — Control plane policy initiative for AKS.
5. Updated ARM Preview API Version Restriction Policy — Advanced policy example with improved exemption logic.
6. AzPolicyLens - Automated Azure Policy Insights — Complementary tooling to manage Azure Policy at scale.

---

This blog post is based on and credits the original work published by Tao Yang at https://blog.tyang.org/2026/08/09/azpolicyfactory-various-updates.