Azure just closed a glaring platform gap: AKS clusters can now mount Azure Files NFS v4.1 shares with Encryption in Transit (EiT) via the Azure File CSI driver. That single change alters threat models for stateful workloads and forces platform architects to rethink three connected things at once storage trust boundaries, cross-tenant identity, and where you run expensive AI inference.
Why this matters now
For years teams treated Azure Files NFS shares as "inside the network" once mounted into a pod: secure enough because the traffic never left Microsoft's backbone. That assumption always had holes multi-tenant clusters, on-prem VPNs, or misconfigured VNET peering could expose NFS RPCs in ways compliance teams hate. EiT for NFS v4.1 via the CSI driver stops that class of surprise by establishing in-transit protection for RPC traffic between node and storage backend.
Technical specifics platform teams need to know
- This targets Azure Files NFS v4.1 and uses the Azure File CSI driver to configure secure NFS mounts. The protection is applied at mount time (RPC security for NFS), not as a generic kernel rewrite of NFS.
- Update and validate your Azure File CSI driver and node images: the driver supplies the mount options and credential handling to negotiate RPC security (for example RPCSEC_GSS/Kerberos-based privacy modes where supported). Confirm your node kernel and distro support the required NFS and RPCSEC features before enabling EiT.
- Treat this like a platform upgrade: test mounts, performance, and failover in staging. Encrypted mounts change latency and error modes for NFS operations in edge cases.
- EiT will impact observability of NFS payloads (encrypted traffic cannot be inspected), so expect changes to monitoring pipelines and DLP integrations.
This is overdue and the consequences are obvious
Platform teams should have enforced encryption in transit for storage long ago. The delay pushed teams into two bad patterns: ad-hoc tunneled mounts and fragile sidecar proxies for file access. Turning on EiT is the right call; the uncomfortable part is the operational churn it creates. Encrypted mounts break packet-level inspection, and any tooling that relied on that will need workarounds or different telemetry.
How this fits with other Azure moves
This AKS change isn't isolated. Microsoft is evolving several areas that together reshape recommended architecture:
- Azure's model-hosting and Foundry offerings are expanding capacity and adjusting pricing, which makes where you host inference a tighter cost vs. capability decision.
- Entra ID (Azure AD) supports federated identity credentials and workload identity federation, which standardizes secret-less cross-tenant authentication; follow Microsoft guidance to migrate legacy issuers and CI/CD identities to those patterns.
- Microsoft is shipping new resiliency and observability capabilities (mix of preview and GA) that nudge teams toward goal-driven resilience and agent-assisted operations.
Put bluntly: encrypted storage, federated identities, and AI-hosting costs are now a single design surface for platform teams building multi-tenant, agentic systems.
Operational implications (short list)
- Identity first: adopt Entra-based federation and workload identity patterns for cross-tenant workloads now. They remove long-lived secrets and simplify multi-tenant auth.
- Separate inference and orchestration: use smaller, cheaper models for routine orchestration and reserve higher-cost hosted models for where capability justifies expense.
- Reinstrument observability: packet-level telemetry will fail on encrypted mounts; rely on application-level metrics, structured logs, and endpoint instrumentation instead.
One candid take
This is a good week for Azure customers but it's also a coordination headache. Microsoft is finally removing footguns (unencrypted NFS traffic) while simultaneously introducing new primitives (cross-tenant federation, shifting model-hosting economics, agent runtimes) that increase the complexity of secure, cost-effective architectures. If your platform team treats these as independent upgrades, you will get slammed by audits, higher bills, or runbook gaps.
If you run stateful, multi-tenant workloads or plan to orchestrate agentic AI on AKS, start with three tasks this sprint: validate CSI driver and node images for EiT, inventory federated identities and issuer usages, and cost-model Foundry inference for your workflows. Ignore any one of those and you'll learn the lesson in production.
Further reading: I covered the AKS NFS EiT change in detail earlier AKS: encryption in transit for Azure Files NFS mounts, and for how Claude landed on Foundry see Claude in Microsoft Foundry on Azure: GA on NVIDIA Blackwell GPUs.
Final thought: platform design in 2026 is convergence. Storage security, identity federation, and AI placement are no longer separate discussions they're the three legs your architecture must stand on.
Sources
- Azure Release Radar – AKS Encryption in Transit for Azure Files NFS via CSI Driver
- Azure Charts – Consolidated Latest Azure Updates (GPT-5.6, Foundry Pricing Update, Chaos Studio CLI)
- Tech Hub – Azure News (Foundry GA updates including GPT-5.6, Asia-Pacific Data Zone, hosted agents)
- Azure Weekly Newsletter Issue 571 – Cohere Command A+, Copilot Observability Agent GA, identity and monitoring retirements
- Azure Service Updates – Infrastructure Resiliency Manager, Azure Migrate + GitHub Copilot, Cosmos DB Synapse Link retirement