CVE-2026-53935: Cilium CiliumLocalRedirectPolicy Namespace Isolation Bypass
Cilium, a widely-deployed networking and security platform for Kubernetes, contains a privilege-escalation vulnerability in how it enforces namespace boundaries. Users who have permission to create CiliumLocalRedirectPolicies can exploit this flaw to redirect traffic destined for Services in other namespaces by manipulating the addressMatcher field. In addition to traffic hijacking, deleting a malicious policy can leave Cilium's internal state corrupted, breaking legitimate service-to-service communication. The vulnerability affects multiple recent versions but has been patched.
Source data · NVD / CISA · public domain
- CVSS
- 3.1 · 6.9 MEDIUM · CVSS:3.1/AV:A/AC:L/PR:H/UI:N/S:C/C:L/I:N/A:H
- Weaknesses (CWE)
- CWE-863
- Affected products
- 0 configuration(s)
- Published / Modified
- 2026-07-07 / 2026-07-10
NVD description (verbatim)
Cilium is a networking, observability, and security solution. Prior to 1.17.16, from 1.18.2 to 1.18.9, and from 1.19.0 to 1.19.3, users with the ability to create CiliumLocalRedirectPolicies can specify arbitrary ClusterIPs via addressMatcher, enabling hijacking traffic to Services in any namespace and bypassing namespace scoping enforced by serviceMatcher; deleting such a policy can also corrupt Cilium internal service state and stop service translation for the affected Service. This issue is fixed in versions 1.17.16, 1.18.10, and 1.19.4.
7 reference(s) · View on NVD →
SEC.co analysis · AI-assisted, reviewed against source
Technical summary
The vulnerability stems from insufficient input validation in CiliumLocalRedirectPolicy's addressMatcher field. While the serviceMatcher parameter correctly restricts policies to services within the same namespace, the addressMatcher field permits arbitrary ClusterIP specification without equivalent namespace scoping. This allows a policy creator to intercept traffic destined for Services across namespace boundaries by matching on their ClusterIP directly. Additionally, policy deletion can trigger corruption of Cilium's in-memory service translation tables, causing the affected Service to become unreachable even after the malicious policy is removed. The root cause is classified under CWE-863 (Incorrect Authorization), reflecting the breakdown in access control enforcement.
Business impact
For multi-tenant Kubernetes clusters, this vulnerability undermines namespace isolation—a fundamental security boundary. Attackers or compromised applications within a namespace can intercept sensitive traffic destined for services in other namespaces, potentially exposing API calls, database connections, or inter-service credentials. The secondary effect of service-state corruption can trigger operational outages, requiring cluster operator intervention and restart of affected components to restore service translation. Organizations relying on namespace-based RBAC policies for tenant separation should treat this as a control-plane integrity issue.
Affected systems
Cilium versions prior to 1.17.16, from 1.18.2 through 1.18.9, and from 1.19.0 through 1.19.3 are vulnerable. Patched versions are 1.17.16, 1.18.10, and 1.19.4 and later. The vulnerability only affects clusters where at least one user or service account has permissions to create CiliumLocalRedirectPolicies; however, in many deployments, cluster operators or platform-engineering teams hold these permissions, expanding the attack surface.
Exploitability
Exploitation requires prior access to create CiliumLocalRedirectPolicy resources within the cluster, typically granted to cluster administrators or platform teams. The attack surface is not remote or unauthenticated—it requires an authenticated principal with RBAC permissions. However, the attack is straightforward once access exists: an attacker simply creates a policy with an addressMatcher targeting any ClusterIP and lets traffic flow through the redirect. This is not listed on CISA's Known Exploited Vulnerabilities catalog, suggesting limited public exploitation to date, though the ease of exploitation means defenders should not rely on low threat-in-the-wild as assurance.
Remediation
Upgrade to patched versions: 1.17.16, 1.18.10, 1.19.4, or any version 1.20 and later. For organizations unable to patch immediately, apply restrictive RBAC policies to limit CiliumLocalRedirectPolicy creation to only highly trusted principals and regularly audit existing policies. Monitor cluster logs and Cilium agent metrics for unusual redirect policies or service translation failures. Ensure namespace isolation is validated post-patching via network policy testing.
Patch guidance
Plan upgrades in a phased manner, testing in staging clusters first, as Cilium upgrades may require brief network-plane restarts. Verify patch versions against Cilium's official release notes to confirm the specific version you deploy contains the fix. After upgrade, confirm service-to-service connectivity across namespaces and review any existing CiliumLocalRedirectPolicies to identify and remove any suspicious entries. Cilium provides vendor advisories with detailed upgrade procedures; consult those for your specific deployment topology (e.g., kubeadm, Helm, managed services).
Detection guidance
Audit existing CiliumLocalRedirectPolicies for addressMatcher values that reference ClusterIPs outside the policy creator's namespace using kubectl get ciliumlocalredirectpolicies -A and inspecting the addressMatcher field. Enable Cilium's audit logging to detect policy creation events. Monitor Cilium agent logs for entries indicating service-state corruption or translation failures, typically logged as warnings or errors in the cilium-agent pods. On older versions, look for any CiliumLocalRedirectPolicy where the Service referenced in serviceMatcher and the ClusterIP in addressMatcher belong to different namespaces—this is a red flag.
Why prioritize this
Although the CVSS score is moderate (6.9 MEDIUM), the vulnerability breaks a fundamental security boundary in multi-tenant clusters. Namespace isolation is a primary trust assumption in Kubernetes security posture; any compromise of that boundary warrants prompt attention. The high-privilege requirement (administrators with policy-creation rights) limits broad blast radius, but many organizations grant these permissions liberally. The corrupted-state side effect creates operational risk. Prioritize patching for clusters with untrusted or semi-trusted users and for those enforcing compliance frameworks (e.g., SOC 2, PCI) that depend on namespace-level isolation.
Risk score, explained
The CVSS 3.1 score of 6.9 reflects a MEDIUM severity: the vulnerability requires administrative or high-privileged access (PR:H) and runs within an adjacent network context (AV:A), limiting immediate remote threat. However, it achieves scope change (S:C) by breaking namespace boundaries and causes availability impact (A:H) through state corruption. The integrity impact (I) is marked none, as the vulnerability does not modify data in transit but does redirect it, a distinction reflected in the scoring. Organizations with strong RBAC discipline may face lower residual risk; those with permissive policy-creation grants should elevate concern accordingly.
Frequently asked questions
Can an attacker exploit this without cluster access?
No. The attacker must first obtain credentials and RBAC permissions to create CiliumLocalRedirectPolicies, typically held by cluster operators or platform teams. Once inside the cluster with those permissions, the attack is trivial.
Does patching immediately restore a corrupted service?
No. After patching to a fixed version, you must restart the affected Cilium agents to reload service state. If service translation remains broken, you may need to delete and recreate the affected Service resource.
How do we test that our cluster is not vulnerable?
Verify your Cilium version matches or exceeds the patched versions (1.17.16, 1.18.10, 1.19.4, or 1.20+). Then audit your CiliumLocalRedirectPolicies to ensure no addressMatcher values reference ClusterIPs from other namespaces. Run a test service in a different namespace and confirm a malicious policy cannot intercept its traffic.
Is this a cluster-wide compromise?
This is not a cluster-wide compromise. It allows traffic hijacking and service-state corruption within the namespace-isolation boundary, but does not grant root access, node access, or the ability to read secrets across namespaces. However, the impact is severe enough to warrant high urgency remediation.
This analysis is based on the CVE description and vendor data current as of the publication date. Patch versions and upgrade procedures should be verified against the official Cilium project releases and security advisories. Organizations should conduct their own risk assessment tailored to their specific deployment topology, RBAC configuration, and regulatory requirements. No exploit code or weaponized proof-of-concept is provided; security researchers and operators are encouraged to test patches in non-production environments before production rollout. Source: NVD (public-domain), retrieved 2026-08-16. Analysis generated by SEC.co (claude-haiku-4-5).
Weaknesses (CWE)
Related vulnerabilities
- CVE-2026-10211MEDIUMAstrBot 4.23.6 Path Normalization Authorization Bypass
- CVE-2026-10616MEDIUMAuthorization Bypass in nextlevelbuilder GoClaw Task Completion
- CVE-2026-10741MEDIUMNexus Repository Manager Credential Disclosure Vulnerability
- CVE-2026-10815MEDIUMAuthorization Bypass in Hostel Management System PHP
- CVE-2026-10860MEDIUMMISP Delete Validation Bypass – Logic Error in HTTP DELETE Handler
- CVE-2026-11379MEDIUMGitLab EE DAST Site Profile Authorization Flaw
- CVE-2026-12352MEDIUMMedium-Severity Authentication Bypass Vulnerability
- CVE-2026-12446MEDIUMChrome Password Manager Cross-Origin Data Leak – Exploit, Patch & Detection