CVE-2026-92574Disclosure

Recovery features often receive less security scrutiny than deployment features, even though they can cross the same privilege boundaries. CVE-2026-92574 is a useful example. Checkpoint restore exists to preserve state, but preserving too much state can preserve authority that the destination environment did not intend to grant. That creates a governance problem as much as a runtime problem. A team may have strong Pod Security controls, carefully reviewed manifests, and restricted namespaces, yet still expose a different path that rebuilds a process with security attributes inherited from elsewhere. **In practical terms, it is a good time to:** - identify every operational workflow that uses CRIU or Kubernetes checkpoint restore for recovery, migration, debugging, or testing - assign explicit ownership for approving checkpoint creation and restore permissions - compare RBAC granted to recovery tooling with RBAC granted to ordinary deployment automation - record checkpoint provenance, source cluster, image digest, and approving identity in centralized logs or change records - test whether restored workloads conform to the destination namespace security policy at runtime Where does checkpoint restore sit in your change-control model: routine deployment, privileged recovery, or something in between? #CyberResilience #Kubernetes #LinuxSecurity #SecurityOperations #Compliance https://linuxsecurity.com/news/security-vulnerabilities/cri-o-restore-kubernetes-privilege-boundaries

5 vulnerabilities that kept security teams busy in October. In October 2026, five foundational infrastructure tools got hit with vulnerabilities that standard monitoring would not have caught: 🔴 #Kubernetes (CVE-2026-92574): security context bypass allowed users with pod creation privileges to break out of container boundaries. 🔴 #Redis (CVE-2026-92925): improper input validation in the cluster bus packet parser enabled remote denial of service and potential code execution. 🔴 #ApacheActiveMQ (CVE-2026-101292): unsafe reflection flaw allowed unauthenticated remote attackers to execute arbitrary shell commands. 🔴 #ApacheAirflow (CVE-2026-82310): token validation failure allowed deactivated accounts to retain indefinite access to pipelines. 🔴 #ApacheZooKeeper (CVE-2026-59969): TLS hostname verification bypass left cluster traffic vulnerable to man-in-the-middle attacks. A green status indicator means the service is running. It does not mean it is secure. 👇 CVE breakdown and patch details in the first comment. #DevOps #DevSecOps #Kubernetes #CyberSecurity #AppSec #OpenSource #Hossted

CRI-O got a nasty one in checkpoint restore: CVE-2026-92574. If someone can create a pod from a crafted checkpoint image, the restored process keeps the checkpoint's UID, capabilities, no_new_privs, and seccomp state instead of the pod's securityContext. Affected upstream is CRI-O 1.34 and later; OpenShift from 4.17 onward. Fixes are in the supported branches (1.34.14 / 1.35.9 / 1.36.6) but not all shipped yet, and remote-registry restore is on by default in affected CRI-O. Treat checkpoint images like untrusted binaries, not just another OCI pull. Registry allowlists and tight create RBAC matter more than the securityContext you wrote in YAML, because admission and audit still report the requested context while the process runs hotter. If you flipped on ContainerCheckpoint or criu for experiments, turn it off until the runtime is patched, and give IR a way to find pods that actually came from restore rather than trusting the YAML alone.

CVE-2026-92574: CRI-O checkpoint restore bypasses destination Kubernetes security context https://www.pruva.dev/reproductions/REPRO-2026-00372 #pruva

🚨 CRI-O CVE-2026-92574 — CVSS 8.8 A pod creator with access to a malicious checkpoint image can bypass the destination Kubernetes security context. Patch pending. → https://threataft.com/articles/cri-o-cve-2026-92574-checkpoint-restore-security-context-bypass #CRIO #Kubernetes #CVE #PatchNow #CyberSecurity #ThreatIntel

CVE-2026-92574 CRI-O checkpoint restore lets a malicious pod bypass destination security context, retaining capabilities and seccomp state for privilege escalation. CVSS 8.8. Patch under review, track it and restrict untrusted checkpoints now. https://www.valtersit.com/cve/CVE-2026-92574 #CRIO #CVEALERT #CVE #infosec #Linux #XSS #valtersit #snippet #SysAdmin #cybersecurity #devsecops #devops #developer #100daysofcode #git #github #gitlab #redteam #blueteam #ethicalhacker #cybersecurityawareness #cybersecuritynews #cybersecuritytips #python #hacker #kali #ubuntu #debian #docker

The kernel enforces the process it receives, not the policy you intended. CVE-2026-92574 matters because checkpoint restore can make those two things differ. **In practical terms, it is a good time to:** - test restore into a stricter destination namespace - inspect NoNewPrivs and capability masks in /proc//status - recreate untrusted restored workloads from normal images #Linux #Kubernetes #AccessControl https://linuxsecurity.com/news/security-vulnerabilities/cri-o-restore-kubernetes-privilege-boundaries

The first vulnerability-management question here is not “What version are we running?” It is “Can anyone actually reach the vulnerable restore path?” CVE-2026-92574 requires checkpoint restore functionality plus permission to create a pod from a malicious checkpoint image. That makes feature use and RBAC part of the exposure calculation, not secondary details. Clusters that never use checkpoint restore should be triaged differently from development or recovery environments where it is enabled. **In practical terms, it is a good time to:** - query cluster configuration for checkpoint/restore enablement and OpenShift ContainerCheckpoint feature-gate use - inspect standalone CRI-O configuration for enable_criu_support - enumerate subjects with create or update access to pods, deployments, daemonsets, statefulsets, jobs, and cronjobs - correlate those permissions with namespaces where checkpoint-based restores are possible #VulnerabilityManagement #Kubernetes #CRI_O #LinuxSecurity https://linuxsecurity.com/news/security-vulnerabilities/cri-o-restore-kubernetes-privilege-boundaries

Pod Security Admission is not a runtime attestation mechanism. It can evaluate what a workload requests, but CVE-2026-92574 highlights a different question: did the process that finally reached the Linux kernel retain the restrictions the policy expected? For restored containers, that distinction matters. A policy-compliant specification does not prove that effective credentials, capabilities, no_new_privs, and seccomp state match the destination configuration. **In practical terms, it is a good time to:** - compare pod securityContext settings with effective process credentials inside restored containers - inspect /proc//status for UIDs, GIDs, capability masks, and NoNewPrivs - verify seccomp confinement for restored workloads rather than assuming admission results imply runtime enforcement - flag checkpoint-based pod creation for additional runtime validation #KubernetesSecurity #AccessControl #Linux #ContainerSecurity https://linuxsecurity.com/news/security-vulnerabilities/cri-o-restore-kubernetes-privilege-boundaries

🚨HIGH - CRI-O Checkpoint/Restore SecurityContext Bypass (CVE-2026-92574) In CRI-O checkpoint restore, restoring a pod from a malicious checkpoint image can cause the process to retain creds, Linux capabilities, no_new_privs, and seccomp state from the checkpoint instead of enforcing the destination Pod securityContext. An attacker with rights to schedule a pod referencing the checkpoint can cross container boundary restrictions and gain elevated privileges. Not exploitable if checkpoint/restore is disabled/unavailable. 👉Affected: cri-o (checkpoint/restore enabled)

A Kubernetes manifest can look locked down while the restored process is running with an older, broader privilege set. That is the uncomfortable part of CVE-2026-92574. CRI-O checkpoint restore can reintroduce credentials, Linux capabilities, no_new_privs, and seccomp state from the checkpoint instead of enforcing the destination pod security context. For platform teams, this turns restore into a trust-boundary problem. Admission policy may approve one configuration while the kernel enforces another. That gap matters during recovery, migration, debugging, and any workflow where checkpoint images cross environments. **In practical terms, it is a good time to:** - inventory clusters and nodes where checkpoint/restore is enabled or has been used - audit Kubernetes RBAC for users and service accounts that can create or update pods and workload controllers - inspect restored workloads and compare effective UIDs, capability sets, no_new_privs, and seccomp state with the intended pod securityContext - restrict checkpoint artifacts to approved registries and controlled storage paths - recreate workloads from normal images if their checkpoint provenance cannot be established How many Kubernetes security reviews validate the process that actually started, rather than only the YAML that requested it? #LinuxSecurity #KubernetesSecurity #ContainerSecurity #DevSecOps #InfrastructureSecurity https://linuxsecurity.com/news/security-vulnerabilities/cri-o-restore-kubernetes-privilege-boundaries

ثغرة CRI-O CVE-2026-92574 (٨٫٨): استعادة حاوية من checkpoint خبيث تقدر تتجاوز securityContext بتاع الـ pod — الـ root والـ capabilities والـ seccomp يرجعوا من الأرشيف مش من الـ spec. لو checkpoint restore شغال عندك، اعتبره حدود صلاحيات مش ميزة. عطّله لحد الباتش، أو قيّد مين يقدر ينشئ pods من صور checkpoint.

CRI-O CVE-2026-92574 (CVSS 8.8): restore from a malicious checkpoint and the destination pod securityContext is theater — root, caps, and seccomp come back from the archive. If checkpoint restore is enabled in your cluster, treat it as a privilege boundary. Disable until patched, or gate who can create pods from checkpoint images.

I once turned on a Tech Preview just to see. Don't do that with ContainerCheckpoint on CRI-O. CVE-2026-92574 restores the checkpoint's root, caps, and seccomp over your pod securityContext (8.8). Fixed in 1.34.14 / 1.35.9 / 1.36.6. Leave the gate off until you can update. https://access.redhat.com/security/cve/cve-2026-92574

There is a new vulnerability with elevated criticality in Red Hat Confidential Compute Attestation and other products (CVE-2026-92574) https://vuldb.com/vuln/408086