AWS Network Policy Agent EUVD-2026-81088

Requires Amazon EKS Network Policy enforcement to be ENABLED (opt-in; if you do not use Network Policy the flaw does not affect you, per the vendor release note) and a running Network Policy Agent before v1.4.0 / VPC CNI before v1.22.4. The attacker must be an authenticated cluster user (PR:H) able to create pods and control pod/namespace names, and must engineer names that produce a pod-identifier collision with a target pod (AT:P) that is co-located and in another namespace. Limiting factors: multi-tenant clusters are the realistic target, namespace/pod naming control is typically privileged, and the collision must line up with an existing victim workload - so it is not exploitable against a single-tenant or Network-Policy-disabled cluster.

This is a genuine but conditional-privilege issue rather than a mass-exploitation threat. The vendor CVSS 4.0 vector (AV:N/AC:L/AT:P/PR:H/UI:N/VC:H/VI:H/VA:N/SC:H/SI:H/SA:N = 8.7) is credible in shape: the high score is driven by the subsequent-system scope metrics (SC:H/SI:H), reflecting that a NetworkPolicy bypass lets one tenant reach pods in OTHER namespaces that should be isolated - a legitimate security-boundary crossing. Importantly, the score is NOT a raw AV:N/PR:N critical: it requires PR:H (an authenticated principal able to create pods/namespaces with attacker-chosen names) and AT:P (a specific pod-identifier collision condition plus a co-located victim pod), which materially raises the real-world bar. There is no EPSS score, no CISA KEV listing, and no public exploit provided, so exploitation status is 'no public exploit identified at time of analysis.' The flaw was self-reported by Amazon (AMZN) with a coordinated patch, and there is no SSVC decision data in the input. Conflicting signal to note: the tag 'Authentication Bypass' overstates the flaw - this is a NetworkPolicy (authorization/segmentation) bypass, not an authentication bypass; VC/VI:H should be read as cross-namespace reachability enabling data access and traffic manipulation, not credential compromise. Net: prioritize for multi-tenant EKS clusters that have Network Policy enforcement enabled, but it is not an internet-facing unauthenticated critical.

In a multi-tenant EKS cluster with Network Policy enforcement enabled, a tenant who can create pods (and influence pod/namespace naming) crafts a pod and namespace whose concatenated pod identifier collides with a victim pod in a different namespace. The collision causes the agent to apply the wrong (or no) NetworkPolicy state, so traffic that should be blocked between namespaces is allowed. No public proof-of-concept is available; exploitation requires authenticated cluster access and engineering a naming collision (AT:P), so it is a targeted, insider/tenant-abuse scenario rather than a drive-by attack.