Releases

AI-analyzed release notes for CNCF graduated and incubating projects.

URGENCY

All Action needed Check if affected Plan ahead

KIND

Category

All AI & ML CI/CD & App Delivery Kubernetes Core Networking & Messaging Observability Orchestration & Management Provisioning & Runtime Security Storage & Data

Aug 2026Clear ×

hamiv2.10.0AI & MLYesterdayAug 21, 2026

ACTION 3CHECK 7OTHER 195View full analysis

A substantial operator-focused maintenance release with broad bug and behavior fixes, alongside new scheduling, device, configuration, and observability capabilities. It also updates security-relevant dependencies and the runtime/toolchain while removing obsolete functionality that may affect compatibility and configuration.

Action needed (3)

securitytensorflow/tensorflow upgraded to 2.21.0rc0-gpu
The tensorflow/tensorflow dependency is upgraded from 2.20.0rc0-gpu to 2.21.0rc0-gpu.
securitytensorflow/tensorflow upgraded to 2.21.0rc1-gpu
The tensorflow/tensorflow dependency is upgraded from 2.21.0rc0-gpu to 2.21.0rc1-gpu.
securitygolang security upgrade
The golang runtime is upgraded to address a security issue.

Check if affected (7)

breakingDRA components removed from the HAMi main chart
Applies if you use DRA components in the HAMi main chart.
breakingWebhook denial of privileged containers
Applies if you use the webhook.
breakingDeprecated scheduler policy configmap removed
Applies if you configure the deprecated scheduler policy configmap.
+ 4 more on the release page

Source ↗

KServev0.20.0AI & MLAug 6, 2026

ACTION 2CHECK 7OTHER 48View full analysis

This release adds new inference-service, storage, runtime, routing, and deployment capabilities, alongside correctness fixes and dependency updates. Operators should review the Starlette security update, changed defaults, and narrower readiness behavior.

Action needed (2)

securitymediumstarlette dependency update for CVE-2026-48710
The starlette dependency is bumped to >=1.0.1 for CVE-2026-48710.
breakingStandard default alignment with the llm-d optimized baseline
The standard default is aligned with the llm-d optimized baseline.

Check if affected (7)

breakinglora-affinity-scorer default for LoRA adapters
Applies if you use LoRA adapters.
breakingLLMInferenceServiceConfig deletion-prevention finalizer
Applies if you use LLMInferenceServiceConfig.
breakingendpointPickerRef default EPP port
Applies if you configure endpointPickerRef and do not configure port.
+ 4 more on the release page

Source ↗

Browse by month