live wire
▸PLATFORM · Clang 23 bytecode interpreter cuts allocations and expands constant-expression supportRed Hat Developer▸PLATFORM · RHEL becomes first-party on OCI; OpenShift subscriptions reach Oracle MarketplaceRed Hat▸AI · vLLM-Omni merges native MOSS-Image serving with parallelism and offloadvLLM▸AI · vLLM adds System One endpoint for Decision 3.0 multimodal routing models (Oct. 15)vLLM▸SECURITY · Red Hat rates OpenShift AI 3.4.5 update Critical; advisory lists hundreds of CVEsRed Hat Product Security▸SECURITY · Red Hat’s first Security Signals report shifts focus from CVE counts to risk paths (Oct. 15)Red Hat▸PLATFORM · IBM adds non-destructive FlexVolume-to-CSI migration tool for OpenShift object storageIBM Cloud▸vLLM tests agent-built Kimi K3 decode path at up to 1.9× speedup; PR remains openvLLM▸AI · LeaderWorkerSet 1.0 changes rollout defaults for disaggregated AI workloadsKubernetes SIGs▸SECURITY · Red Hat flags active exploitation of kernel flaw affecting OpenShift CoreOS; no fix listedRed Hat Product Security▸AI · Red Hat sets Dec. 9 showcase for AI 3.6 agent governance and model servicesRed Hat▸vLLM 0.32.0 drops Python 3.10, restricts CORS defaults and adds security fixesvLLM▸SECURITY · Red Hat tracks Moderate Hono auth bypass across OpenShift AI, Dev Spaces and AnsibleRed Hat Product Security▸SECURITY · Red Hat Hardened Images fixes Moderate libpng denial-of-service flawRed Hat▸PLATFORM · Clang 23 bytecode interpreter cuts allocations and expands constant-expression supportRed Hat Developer▸PLATFORM · RHEL becomes first-party on OCI; OpenShift subscriptions reach Oracle MarketplaceRed Hat▸AI · vLLM-Omni merges native MOSS-Image serving with parallelism and offloadvLLM▸AI · vLLM adds System One endpoint for Decision 3.0 multimodal routing models (Oct. 15)vLLM▸SECURITY · Red Hat rates OpenShift AI 3.4.5 update Critical; advisory lists hundreds of CVEsRed Hat Product Security▸SECURITY · Red Hat’s first Security Signals report shifts focus from CVE counts to risk paths (Oct. 15)Red Hat▸PLATFORM · IBM adds non-destructive FlexVolume-to-CSI migration tool for OpenShift object storageIBM Cloud▸vLLM tests agent-built Kimi K3 decode path at up to 1.9× speedup; PR remains openvLLM▸AI · LeaderWorkerSet 1.0 changes rollout defaults for disaggregated AI workloadsKubernetes SIGs▸SECURITY · Red Hat flags active exploitation of kernel flaw affecting OpenShift CoreOS; no fix listedRed Hat Product Security▸AI · Red Hat sets Dec. 9 showcase for AI 3.6 agent governance and model servicesRed Hat▸vLLM 0.32.0 drops Python 3.10, restricts CORS defaults and adds security fixesvLLM▸SECURITY · Red Hat tracks Moderate Hono auth bypass across OpenShift AI, Dev Spaces and AnsibleRed Hat Product Security▸SECURITY · Red Hat Hardened Images fixes Moderate libpng denial-of-service flawRed Hat
upstreambeat.ai
newsAI

Red Hat AI Inference 3.5 extends MaaS governance to AKS

The Kubernetes deployment adds API-key authentication, subscription-aware access and token quotas in front of vLLM and KServe model endpoints.

AKS AI gateway with policy checks versus raw model access
AI-generated illustration
By The News Desk· Oct 13, 2026the quick take — two AI hosts go live when you do

Red Hat has documented how the Model-as-a-Service governance layer in Red Hat AI Inference 3.5 runs on Azure Kubernetes Service, extending governed model access beyond OpenShift. The engineering walkthrough shows MaaS installed above a vLLM, llm-d and KServe inference stack on a four-node AKS cluster.

What the governance layer adds

A raw KServe inference gateway can expose an OpenAI-compatible model endpoint without authentication or rate limits. The MaaS deployment adds a second, policy-enforced gateway. Requests pass through Authorino for API-key or JWT validation and Limitador for token-rate checks before reaching the model backend.

Red Hat Connectivity Link, the productized Kuadrant layer, supplies those enforcement engines. A MaaS controller translates higher-level MaaSSubscription and MaaSAuthPolicy resources into Kuadrant authentication and token-rate-limit policies. A separate MaaS API issues, validates and revokes API keys; its PostgreSQL backend stores key hashes and subscription metadata rather than plaintext keys.

The test demonstrated the expected outcomes: a request without a key returned HTTP 401, a request with a valid key returned a model response, and requests beyond a 1,000-token-per-minute subscription limit returned HTTP 429. The gateway blocks rejected traffic before it consumes model compute.

What changes outside OpenShift

AKS does not provide the same integrated installation, certificate and identity facilities as OpenShift. The documented deployment therefore uses Helm rather than OpenShift's operator installation path, cert-manager for TLS and an external OpenID Connect provider such as Microsoft Entra ID. An operator chain selects AKS-specific manifests and deploys the MaaS controller, API, gateway and policy resources.

The model-serving path remains separate. In the example, one LLMInferenceService resource deployed Qwen3-0.6B through vLLM on an NVIDIA T4 node, while the MaaS components ran on CPU. Platform teams can keep direct inference access available or send consumers through the governed gateway.

Why it matters

The notable change is not simply that Red Hat's inference runtime works on managed Kubernetes. It is that the access-control and quota layer now follows it. That gives teams running AKS a Kubernetes-native way to issue model API keys, bind users or groups to models, enforce per-subscription token budgets and use the same policy path for locally hosted or external models.

The article is a deployment walkthrough, not performance or scale evidence. Teams evaluating it should still test identity integration, key rotation, gateway availability and quota behavior under their own traffic before treating the configuration as a production service.

sources

Filed by The News Desk. Corrections: desk@upstreambeat.ai · Our standards →

comments · 0

    Comments are moderated before they appear. Your email is used once to confirm it is you — never shown, never sold. Corrections and questions get an answer from the desk when we have one.

    more from the desk