Zero Trust in Practice: Avoiding YAML Pitfalls with Kubernetes NetworkPolicy and Istio
A practical demonstration of Zero Trust principles in Kubernetes reveals how a single character error in a NetworkPolicy manifest can silently open access to all Pods instead of enforcing strict ingress rules.
The scenario involves an API running in the shop namespace that must read orders from the orders service in the backend namespace over TCP port 8080. Simply being inside the cluster must not grant neighboring services the same access. The policy named orders-from-api is intended to restrict connections, while orders-mtls enforces mTLS and orders-read-from-api further limits HTTP methods and paths.
NetworkPolicy objects can exist in the API server without activating any filtering because enforcement depends on the CNI plugin. Before applying rules, administrators must verify CNI support and test Pods using hostNetwork separately. Commands such as kubectl -n backend get pods -l app=orders --show-labels help confirm that the correct Pods are selected by label.
The correct policy requires both namespaceSelector and podSelector conditions to be satisfied together inside the same from element:
- namespace must match kubernetes.io/metadata.name=shop
- pod must match app=api
Adding an extra hyphen before podSelector creates a second independent from element, allowing any Pod from the shop namespace or any Pod labeled app=api from the backend namespace. This mistake was detected only when additional test Pods were introduced.
Testing occurred on a single-node K3s v1.36.5+k3s1 cluster running Alpine 3.24.2 with four vCPU and 4 GB RAM. The orders service, built with BusyBox 1.37.0, served a simple lab-ok response. Results showed that only the intended client received HTTP 200 while others were refused when the policy was correctly written.
After network restrictions, Istio was introduced to enforce mTLS and application-layer controls. A PeerAuthentication resource named orders-mtls set mode STRICT for Pods labeled app=orders. An AuthorizationPolicy named orders-read-from-api then permitted only GET requests to /orders/* from the principal cluster.local/ns/shop/sa/orders-api.
Verification on a separate K3s cluster with Istio 1.31.3 minimal confirmed that only the correctly identified client could issue successful GET requests, while POST requests and requests from other identities were rejected with appropriate HTTP status codes or connection resets.
Finally, the study examined who can create Pods using the orders-api ServiceAccount. RBAC checks with kubectl auth can-i create pods revealed that an operator could delegate the ServiceAccount identity, allowing the created Pod to inherit its permissions. This risk requires additional admission controls and restrictions on spec.serviceAccountName.
Related articles
Home Lab Overhaul: Securing Proxmox Infrastructure After Discovering 63,000 Forgejo Spam Accounts
A home lab administrator migrated Docker stacks to new LXC containers on Proxmox and uncovered multiple security misconfigurations. The setup relied on Cloudflare proxying, MikroTik port forwarding to NPMplus with CrowdSec, and Cloudflare Tunnels for admin panels. After fixing IP address logging in the WAF, logs revealed 63,021 spam accounts created on a Forgejo instance that had open registration. The operator disabled registrations, recreated the instance from verified Git mirrors, and eliminated direct internet exposure by routing all traffic through Cloudflare Tunnel. Additional controls included Zero Trust policies, device enrollment restrictions, and TOTP on Guacamole. The changes removed open ports, enforced proper client IP visibility for CrowdSec, and segmented public, limited, and admin access paths.
Benchmarking AES-GCM and Ascon AEAD Performance on ESP32 and ESP8266 for Industrial Telemetry
Researchers conducted detailed experiments measuring the real-world costs of three AEAD implementations on ESP32 and ESP8266 microcontrollers used in industrial telemetry scenarios. The study compared mbedTLS AES-128-GCM, legacy Ascon-128 via CryptoLW, and a portable C++ port of NIST Ascon-AEAD128 across three load profiles ranging from stable low-frequency packets to maximum-rate 10 ms intervals. Key metrics included AEAD execution time, CPU cycles, memory usage, UDP transmission success rates, and power consumption captured via TC66C. Initial tests revealed significant UDP packet loss on ESP32 under MAX load due to ENOMEM and ENOBUFS errors in the lwIP stack, which were largely resolved after network handling improvements. Results showed AES-GCM completing in roughly 618 µs on ESP32 versus 803 µs for legacy Ascon-128, while the reverse held on the slower ESP8266 where legacy Ascon outperformed AES by a factor of approximately 3.5. The work highlights that network stack limitations often dominate over cryptographic overhead in high-throughput industrial sensor deployments.
ARinteg Earns Golden Partner Status from Ideco for NGFW Security Projects
ARinteg has been awarded Golden Partner status by Ideco following successful joint projects in network security. The integrator supplies and implements the Ideco NGFW Novum next-generation firewall for perimeter protection and secure remote access. Collaboration also covers customer migrations from foreign NGFW solutions to the Russian vendor’s product. Commercial Director Dmitry Slobodenyuk stated the new status will allow ARinteg to deliver fuller support across all implementation stages. Ideco’s Partner Relations Head Sergey Markin highlighted ARinteg’s technical expertise in large-scale deployments. The partnership supports import substitution and hybrid infrastructure requirements.
Preventing Secret Leaks: Integrating Secret Managers with Automated Scanners
The article explains how organizations can move from reactive secret detection to proactive prevention by combining Secret Managers with automated scanners. It breaks down the full lifecycle of a secret across four stages: creation, storage, usage, and rotation. Common failures include manual generation of long-lived static tokens, storage in Git repositories or unencrypted files, and lack of automated rotation. Solutions center on dynamic secrets via tools like Deckhouse Stronghold and HashiCorp Vault, paired with scanners such as CodeScoring.Secrets and Gitleaks that enforce checks at commit and CI stages. The piece also covers runtime protections including in-memory injection for Kubernetes and secret masking in logs. Real-world examples highlight risks from forgotten service tokens and hardcoded credentials that persist for years.