AKS just closed a gap that’s been quietly gnawing at platform teams: general support for Encryption in Transit (EiT) on Azure Files NFS v4.1 mounts via the Azure File CSI driver.
This matters because stateful workloads often depend on NFS mounts for databases, caches, or legacy storage layers, and until now you had two uncomfortable choices: accept cleartext NFS inside your cluster and bolt on network isolation, or invent a host-level proxy/tunnel layer that added operational complexity and failure modes. With EiT available in recent Azure File CSI driver releases, AKS teams get a supported, auditable path to satisfy stricter compliance baselines and reduce custom plumbing.
Why this is not just checkbox security
Encryption at rest and TLS for node‑to‑storage were table stakes in many vendor stacks — but NFS historically exposed an internal data plane that most auditors flagged. A TLS-enabled CSI mount changes the attack model: the NFS session is cryptographically protected between the pod (via the CSI driver) and the Azure Files service. That reduces the utility of lateral network sniffing inside a kube network and shrinks the list of required compensating controls.
That said, EiT is necessary, not sufficient. Toward the operational implications:
- Verify your Azure File CSI driver release supports EiT and add the explicit mount/volume attribute required to enable TLS in your PersistentVolume or StorageClass definitions.
- You still need to consider node compromise: if a node or kubelet is compromised, an attacker with root can access mounted volumes. TLS protects the wire; it doesn't change host-level trust assumptions.
- Performance: TLS introduces CPU and handshake overhead; test throughput on typical NFS workloads (databases, backup jobs) before rolling it cluster-wide.
If you want a quick refresher on the AKS work here, we previously covered similar changes: AKS: encryption in transit for Azure Files NFS mounts.
The rest of the week: identity, resilience, and agentic AI
The storage update pairs with other non-sexy but consequential platform moves. Managed Identities gained more federated cross-tenant scenarios, and Microsoft is tightening workload identity flows — moving toward Microsoft Entra/OIDC federation for CI/CD and workload identities. This centralizes federated identity and standardizes secretless auth across tenants and CI/CD systems. If your pipelines still inject long-lived secrets into runners, this week should make you uncomfortable.
Bicep also shipped a plan mode — a change preview without deployment — and Azure introduced an Infrastructure Resiliency Manager in public preview. Together they let platform teams simulate changes and validate resiliency against failure scenarios before hitting prod. Bicep plan finally gives teams a supported dry-run instead of fragile hacks; Infrastructure Resiliency Manager formalizes failure hypotheses and reduces risky deploy-and-roll-back cycles that drive costs and outages.
And yes, Azure AI/Foundry kept expanding the frontier. Foundry added larger and mixture‑of‑experts model options and pushed more agentic capabilities into the Foundry Agent Service. The practical bit: these models are tuned for low‑latency, multimodal, and agentic workflows — they’ll be glued into orchestration layers and observability pipelines. That makes the identity and storage changes above relevant: agentic workloads will be accessing data stores, invoking downstream APIs, and performing actions across tenants.
Opinion: the platform trade-offs are finally being acknowledged
This week feels like plumbing — but it's the plumbing that bites you when you ship an agent that can read a mounted NFS share and make outbound calls. Enabling EiT was overdue; platforms needed a supported way to close the NFS wire. Equally, converging identity on Entra and adding Bicep plan mode are the right calls. Microsoft is shifting from permissive defaults and ad-hoc patterns toward auditable, federated primitives. Teams that treat these as optional "nice-to-haves" will be the ones doing expensive incident postmortems when agentic models meet legacy storage.
If you run stateful workloads on AKS, enable EiT in your test clusters this sprint, benchmark the overhead, and treat workload identity migration as a roadmap item, not a future task. These bits don't make exciting release notes, but they materially change what you can safely automate with agents and who you trust with your pod‑mounted data.
Sources
- Azure Kubernetes Service – Encryption in Transit for Azure Files NFS v4.1 volumes
- Azure Infrastructure and SQL platform updates (Azure Service Updates – Robin Konrad)
- Azure Weekly Newsletter Issue 571 – AI, observability, identity, Bicep, and retirement notices
- Announcements – Microsoft Azure Blog (agentic AI and Foundry-related announcements)
- News – Azure – Tech Hub (Foundry GPT‑5.6 GA, Asia-Pacific Data Zone, hosted agents)