Azure just made it practical to run unmodified Linux containers inside hardware-backed confidential VMs in AKS — GA. That sentence contains the operational pivot: AKS now offers a lift‑and‑shift path into confidential computing with VM-level isolation and encryption baked into the cluster control plane. You don’t need enclave-aware code or special CRDs; you need to treat the node as a stronger trust boundary and rethink how telemetry and access cross it.
This isn’t an academic checkbox. Confidential VMs for Azure Linux in AKS mean platform teams can move PII-handling, crypto, and other sensitive workloads into TEE-backed VMs (the usual suspects: AMD SEV, Intel TDX families) while keeping standard container deployment workflows. That’s the right move — hardware-backed isolation belongs at the platform layer, not in bespoke app rewrites — but it exposes immediate operational questions: how do you get logs, metrics, and traces out of a VM you’ve intentionally made opaque to operators?
Microsoft’s other announcements this week answer that tension — inconsistently. On the one hand, Azure Monitor added tighter isolation features for workspaces, introduced resource-scoped queries for limiting scope, and published lower-cost retention/ingestion tiers for auxiliary logs. These changes can reduce blast radius and lower bills for high-throughput telemetry, if you reorganize where high-volume, low-value logs land.
On the other hand, Microsoft published deprecation and migration guidance for several legacy paths. Legacy authentication flows for Container Insights and older container telemetry schemas are being deprecated; AKS-Engine has been deprecated for years and teams are being encouraged to migrate to managed AKS. Microsoft is urging migration of container telemetry to the newer ContainerLogV2 schema and to review the timelines published in their docs for your tenant.
Telemetry is the hidden cost
If you plan to use confidential nodes, map your log and agent topology now. ContainerLogV2 changes how container stdout/stderr and structured logs are stored and queried; schema migrations will break dashboards and alert rules that assume the old table. Azure’s lower-cost tiers for auxiliary logs help, but only if you re-architect ingestion to push high-volume, low-value logs into auxiliary streams and keep security-critical events in primary tables. Resource-scoped queries are welcome — they let you limit workspaces to the minimal scope — but they also make cross-tenant diagnostics harder unless you build a controlled aggregation layer.
Foundry and frontier models: region-aware data zones
Microsoft Foundry announced availability of higher-capacity frontier-class models in region-specific data zones and new agent capabilities. Practically, that means larger LLM instances hosted in regional data zones and more productized agent tooling across observability. The Copilot Observability Agent and related offerings now have pricing and table/retention plans that tie agent behavior to Monitor table strategies; that coupling accelerates adoption and creates a new cost surface to watch.
One honest take: Microsoft is doing the right thing by making confidential computing accessible at the AKS node level. Platform teams should prefer an integrated VM-based trust boundary over shoehorning enclaves into app code. But Microsoft is also moving quickly with telemetry primitives — deprecations and the enforced shift to ContainerLogV2 will create breakage for teams that treat monitoring as "plumb it and forget it."
Operationally, treat the next three months as a single project: (1) decide which clusters will run confidential nodes, (2) map agents and logging flows for those clusters and migrate schemas to ContainerLogV2, and (3) adopt resource-scoped Monitor configurations and admin APIs for workspace networking so you don’t leak sensitive telemetry outside your tenant box. If you want a thread on Foundry’s trajectory, this builds on Microsoft’s earlier Foundry work — region-aware frontier models plus agentic tooling is now a platform pattern, not a beta.
Prediction: within 12–18 months platform engineering teams will treat node confidentiality as a first-class architectural decision that shapes telemetry pipelines. If you haven’t tested schema migrations and resource-scoped queries before your tenant’s deprecation timelines, you’ll be doing them under fire. The good news: the primitives are here. The bad news: Azure just gave you a stronger trust boundary — and a migration deadline.
Sources
- Azure Product Updates Digest – AKS Confidential Virtual Machines GA (September 1, 2026)
- Microsoft Azure Blog – GPT-5.6 now available in Microsoft Foundry
- Azure Updates – Azure Copilot Observability Agent supports Basic and Auxiliary table plans; Azure Firewall auto-learn SNAT routes
- Azure Update Storyline – Azure Monitor network security perimeter GA, auxiliary logs reduced pricing, ContainerLogV2 and legacy auth retirements
- Tech Community – Check This Out! (CTO) Guide (September 2026)
- Azure Updates Digest – Retirement of Azure Network Policy Manager for AKS Windows nodes by September 30, 2026