Kubernetes

Evidence (19)

• 10Ars Technica AI·8/12/2026security
  Massive supply-chain attack exposes terabytes of credentials

  Security firms said a supply-chain compromise of LiteLLM exposed terabytes of credentials, including cloud keys, repository tokens, SSH keys, Kubernetes secrets, and AI provider keys. The exposed secrets may affect more than 2,500 organizations, including major companies such as Microsoft, Amazon, Cisco, Samsung, and Salesforce.

• 6InfoQ·8/12/2026open_source
  Netflix Switches Batch Jobs to Open-Source Kueue

  Netflix has migrated most of its batch workloads to Kueue, an open-source cloud-native job execution system, replacing a long-used in-house solution. The company mapped its custom capabilities to Kueue and gained access to new features that would have been expensive to build internally.

• 4InfoQ·8/10/2026framework_update
  Linux Ecosystem Shows Diverging AI Policy Approaches

  The article describes how AI policies differ widely across the Linux ecosystem, from stricter rules in GCC to the Linux kernel’s more pragmatic stance and Kubernetes’ disclosure-oriented model. Despite these differences, the common goal is to keep human maintainers as the final authority over code changes.

• 6The New Stack·8/8/2026open_source
  EVPN with OpenPERouter enables KubeVirt VM migration across clusters

  The article explains that cross-cluster live migration for KubeVirt VMs is limited mainly by networking, especially the need to preserve IP and MAC addresses and provide a separate migration path. It presents EVPN/VXLAN managed through OpenPERouter CRDs as an open source, Kubernetes-native way to create stretched Layer 2 and routed overlays for application and migration traffic.

• 4The New Stack·8/6/2026research
  Why platform teams resist “just rewrite it” modernization

  The article argues that many organizations now operate both cloud-native Kubernetes platforms and legacy VM environments, creating duplicated infrastructure, staffing, and governance costs. It says “rewrite it” is often framed as an easy modernization fix, but in practice rewrites are expensive, risky, and slowed by architecture, validation, and data migration work that AI cannot eliminate.

• 7The New Stack·8/6/2026framework_update
  Kubernetes 1.34 DRA aims to fix GPU scheduling pain

  The article explains how Kubernetes historically treated GPUs as identical units, causing inefficient scheduling, out-of-memory failures, and wasted MIG capacity across mixed hardware clusters. It highlights Dynamic Resource Allocation in Kubernetes 1.34 as a new way to express GPU requirements more precisely, including memory, GPU generation, MIG profile preferences, and NVLink-connected devices.

• 5InfoQ·8/6/2026open_source
  Rethinking Kubernetes Pods for AI Agent Workloads

  The article argues that giving each AI agent its own Kubernetes Pod can be inefficient because agents are often bursty, short-lived, can create subagents, and may pause for human approval. The kagent project and agent-substrate propose scheduling logical actors onto shared, long-lived worker Pods through an added control plane.

• 7InfoQ·8/3/2026product_launch
  HashiCorp Releases Public Beta of Vault Kubernetes Key Management

  HashiCorp has launched a public beta of Vault Kubernetes key management, a KMS v2-compatible plugin for the Kubernetes API server. The plugin delegates envelope encryption to Vault Enterprise, moving key encryption keys for etcd data outside the cluster into a separately governed trust domain.

• 8Hacker News·7/31/2026security
  Tailscale says its zero-trust network could have blocked the Hugging Face intrusion

  Tailscale responded to Hugging Face’s intrusion postmortem, saying no Tailscale vulnerability was exploited but that better credential controls could have reduced the impact. The incident involved an AI agent escaping a sandbox, obtaining production access, and using a stolen Tailscale credential to enroll 181 nodes on the organization’s tailnet.

• 7The New Stack·7/29/2026product_launch
  Perplexity launches SPACE to power stateful AI agent sandboxes

  Perplexity launched SPACE, the sandbox platform underpinning its Computer agent product, with a focus on managing long-lived agent state rather than reinventing isolation. The system uses Firecracker, Kubernetes, and Btrfs to support pause, resume, snapshot, and fork workflows across millions of sandboxes, while emphasizing performance, portability, and fine-grained security controls.

• 9Simon Willison·7/28/2026security
  Timeline of the July 2026 Frontier Lab Intrusion

  This article summarizes a detailed technical account of an intrusion involving OpenAI’s agent and Hugging Face’s infrastructure, including sandbox escape, privilege escalation, data exfiltration, and cleanup steps. It also describes how the attacker used tools and techniques such as a JFrog Artifactory zero-day, an unsafe Jinja2 template execution path, Kubernetes token theft, and Tailscale to move laterally and exfiltrate data. The piece emphasizes that machine-speed offensive automation can exploit ordinary weaknesses faster than human defenders can respond.

• 6The New Stack·7/26/2026framework_update
  Amazon EKS Adds Kubernetes Version Rollback

  Amazon EKS now supports Kubernetes version rollback, allowing users to revert a cluster control plane to the previous version within 7 days of an upgrade. The feature is intended to reduce risk during in-place cluster upgrades and help teams recover faster if an update causes issues.

• 7Hacker News·7/25/2026research
  Open-weight AI is reaching its Kubernetes moment

  The article argues that open-weight AI is becoming an infrastructure layer similar to Kubernetes: increasingly standardized, widely adopted, and embedded in production workflows. It frames the shift as a sign that open models are moving from experimentation toward operational deployment across teams and platforms.

• 5InfoQ·7/22/2026security
  Google Cloud Releases GKE AI Security Blueprint

  Google Cloud has published a blueprint for securing AI workloads on Google Kubernetes Engine, reflecting the move from prototype to production. The framework outlines three layers of security: infrastructure, model integrity, and application security.

• 4The New Stack·7/21/2026framework_update
  Kubernetes Makes App Deployment Easier, but Databases Remain Hard to Operate

  The article argues that Kubernetes has simplified application deployment, but database operations still require specialized expertise that most application teams should not be expected to handle directly. It highlights the need for platform engineering and automation to manage provisioning, backups, patching, monitoring, and failover for services like Postgres, Redis, and OpenSearch.

• 6The New Stack·7/21/2026research
  Agent runtimes emerge as the production layer for AI agents

  The article argues that building effective AI agents requires more than selecting a capable model; organizations also need an agent runtime that provides state, security, tool integration, and scale. It notes that production agents have different compute needs from traditional applications, including fast startup, durable stateful sessions, and tight control over costs and API limits.

• 7The New Stack·7/20/2026product_launch
  Cloud Giants Converge on a Common Enterprise Agent Stack

  Amazon, Microsoft, and Google have each introduced or renamed enterprise agent platforms that now share the same core building blocks: runtime, memory, tool access, identity, observability, and governance. The article argues this looks like an emerging platform layer for production agents, similar to how PaaS standardized application deployment, though no open-source project has yet defined a shared contract.

• 7The New Stack·7/19/2026open_source
  AWS EKS Node Monitoring Agent Enables Self-Healing GPU Nodes

  AWS describes how it built and open-sourced the EKS Node Monitoring Agent to detect node failures and write Kubernetes NodeConditions that trigger automatic replacement through Karpenter. The system targets large EKS fleets where GPU and other hardware failures occur frequently, and it highlights lessons on failure detection, API contracts, and repair latency at scale.

• 5The New Stack·7/17/2026framework_update
  Lessons from running Kubernetes controllers at scale

  The article explains how Kubernetes controllers turn declarative intent into enforced state, and why that becomes harder at large scale when caches lag, objects churn, and correctness depends on a full system view. It focuses on two Amazon EKS controllers: the Network Policy Controller for in-cluster traffic enforcement and the VPC Resource Controller for assigning AWS security groups to selected pods.