AKS just closed a long-standing storage security gap: Encryption in Transit for Azure Files NFS shares reached general availability on July 17. For platform engineers running stateful workloads that mount NFS-backed Azure Files into pods, this is not a minor checkbox — it changes how you design networking and threat models around persistent volumes.
What changed, practically
Before this GA, teams that needed confidentiality for data sent to Azure Files over NFS often fell back to network-level encryption (IPsec tunnels, VPNs, or VNet peering) or application-layer encryption inside the workload. Those are brittle: IPsec adds operations and debugging overhead, peering requires specific topology, and application encryption passes the storage responsibility back to developers. GA for encryption in transit removes that middle step — the platform now offers an encrypted channel for NFS mounts from AKS nodes to Azure Files.
That has three immediate implications for platform teams:
- It shrinks the attack surface for exposed nodes. You don't have to rely on overlay tunnels to protect NFS traffic crossing less-trusted network segments.
- It simplifies compliance posture for workloads that required encrypted-in-transit storage but couldn't tolerate the latency or ops cost of application-level encryption.
- It changes CSI and mount configuration decisions: the Azure Files CSI driver and the storage class options you choose now have security consequences that the platform can enforce, instead of relying on custom init containers or sidecars.
Don't misread this as "set it and forget it." Encryption in transit is necessary but not sufficient. RBAC on the CSI driver, key management for storage accounts, and audit logging for mount activity still need platform-level guardrails. If your clusters use cross-tenant or hybrid connectivity, double-check how the encryption mode interacts with your VPN or ExpressRoute configuration.
Microsoft's AI push is the other headline
Alongside the storage update, Microsoft has been expanding Foundry and related AI tooling: providers such as Anthropic are accessible through Microsoft's Foundry marketplace and Microsoft is shipping more agent runtimes and preview agent services. That matters because these agent runtimes and model services will create more managed processes and sidecars that store checkpoints, logs, and model artifacts nearby or on mounted storage.
Stronger storage guarantees plus more agent-first AI workloads is an important combination. Agent runtimes will need secure, mounted storage for checkpoints, logs, and model artifacts. Now that AKS can encrypt NFS mounts in-transit, teams will be tempted (and rightly so) to use Azure Files for agent state. But that also means platform teams must manage secrets, quotas, and observability for model agents at cluster scale. The operational surface has widened.
A note on networking: IPv6 for VPN Gateway
In parallel, Azure updated IPv6 support for VPN Gateway and related networking components. Native or dual-stack IPv6 on VPN endpoints simplifies hybrid topologies and reduces NAT workarounds. For AKS clusters connected to on-prem or edge via VPNs, this reduces some network-level hacks and pairs well with in-transit encryption: dual-stack connectivity plus encrypted mounts means fewer manual network protections to secure storage traffic.
Why platform teams should act
If your AKS workloads mount Azure Files NFS shares, treat this GA as an operational requirement, not an optional improvement. Revisit storage classes and the Azure Files CSI driver configuration, retire brittle IPsec or tunnel workarounds where appropriate, and bake encryption-in-transit into your platform defaults. Simultaneously, inventory any emerging Foundry/agent workloads for storage needs, secrets, and observability — the new AI surface will consume storage and network capacity in ways that teams focused only on CPU and GPU quotas may miss.
Microsoft is doing the right thing by hardening a basic building block (NFS mounts) while pushing higher-level AI primitives. The part that will bite teams is the interaction: encrypted storage is meaningless without platform controls for agents and telemetry. If you ignore that second half, you'll have secure mounts—and model agents that silently burn through storage, egress, and budget.
For quick next steps, see our concisely focused writeup on the feature AKS: encryption in transit for Azure Files NFS mounts and read how Foundry's model and agent pushes are already changing platform responsibilities in Microsoft Foundry Adds Anthropic Claude (GA), GPT-4o Support, and Foundry Agent Service GA.
Azure is nudging platform teams toward a single truth: storage, networking, and AI runtimes are converging operationally. If you don't centralize control now, you'll be chasing fractured fixes later.