Submariner CVE-2026-66785

Requires the attacker to control a malicious spoke cluster that is already joined and trusted as a participant in a Submariner multi-cluster network, and the deployment must use Submariner cross-cluster connectivity within Red Hat Advanced Cluster Management. Exploitation hinges on the system's failure to validate subnet/endpoint advertisements (CWE-20), so a peer cluster will accept attacker-declared arbitrary network ranges. Limiting factors: the attacker needs a privileged, authenticated foothold as a mesh member (reflected in PR:H) rather than anonymous network access; the vulnerability is only reachable where Submariner inter-cluster networking is enabled and multiple peer clusters are connected; no victim user interaction is required.

This is a genuine but conditional priority rather than a drive-by internet threat. The CVSS 3.1 vector (AV:N/AC:L/PR:H/UI:N/S:C/C:H/I:H/A:N, base 8.7 per Red Hat) tells a consistent story: network-reachable and low-complexity to trigger, but gated behind PR:H - the attacker must already control a federated 'spoke' cluster inside the Submariner mesh, which is a privileged position of trust. The scope change (S:C) and high confidentiality/integrity impact are well justified because a malicious cluster can reroute peer clusters' traffic through its own tunnel, affecting systems beyond the attacker's own cluster. The exploitation signals are more reassuring: EPSS is only 0.29% (21st percentile), there is no CISA KEV listing, and no public exploit code is referenced - so there is no public exploit identified at time of analysis. The low EPSS is consistent with the high barrier to entry (you need to be an authenticated participant in a multi-cluster fabric), not with a widely-sprayed vulnerability. No SSVC decision data was provided in the input, so exploitation-status/automatable/technical-impact cannot be cited from a vendor SSVC record. Net: treat as a high-severity trust-boundary/tenant-isolation flaw to patch promptly in multi-cluster Submariner deployments, but it is not an unauthenticated internet-facing emergency.

An attacker who controls a spoke cluster already joined to the Submariner multi-cluster network publishes a crafted endpoint advertising arbitrary/overly-broad subnets that the system fails to validate. Peer clusters accept the advertisement and begin routing traffic destined for those ranges through the attacker's tunnel, enabling interception (information disclosure) or blackholing/disruption. No public exploit code is referenced, and exploitation requires an existing privileged foothold as a mesh participant (PR:H).