Red Hat ACM CVE-2026-66787

Exploitation requires the attacker to hold sufficient Kubernetes RBAC privileges on a spoke cluster to create or modify EndpointSlice objects - corresponding to the CVSS PR:H rating. This implies the attacker has already compromised the spoke cluster at a cluster-admin or at minimum an endpoint-writer service account level. The ACM multi-cluster hub must be configured with lighthouse-based cross-cluster service discovery active and the compromised spoke cluster must be joined and trusted by the hub. Exploitation is limited to environments where cross-cluster service communication is in active use; single-cluster or non-lighthouse-connected ACM deployments are not affected. The attack does not require any user interaction on the victim clusters and operates transparently once the malicious EndpointSlice is accepted.

The CVSS base score of 8.7 reflects the Scope:Changed characteristic (S:C), which elevates severity because a single compromised spoke cluster can affect all participating clusters in the ACM fleet - not just the attacker's own environment. However, PR:H is a critical limiting factor: exploitation requires an attacker who already controls a spoke cluster at a sufficient privilege level to create or modify EndpointSlice objects. This positions the threat actor as an insider, a supply-chain compromise of cluster workloads, or a post-exploitation pivot within an already-breached cluster. No EPSS score was provided in the source data, so exploitation probability cannot be quantitatively assessed. No KEV listing exists, and no public POC has been identified, indicating this is not currently being broadly exploited in the wild. The realistic risk profile is: high impact if a spoke cluster is already compromised, but the PR:H prerequisite significantly narrows the population of viable attackers. Organizations running ACM with many spoke clusters should treat this as a high-priority internal lateral movement and persistence risk rather than an external initial-access vector.

An attacker who has gained administrative access to a spoke cluster - via credential theft, a compromised CI/CD pipeline, or exploitation of another vulnerability - creates EndpointSlice objects with IP addresses pointing to attacker-controlled infrastructure. Lighthouse DNS on the hub and peer clusters begins resolving the targeted cross-cluster service names to the malicious IPs, transparently redirecting application traffic. The attacker acts as a relay, forwarding traffic to the legitimate backend while silently reading or modifying the payloads, with no indication to the communicating applications that interception is occurring.