Lighthouse CVE-2026-66788

Exploitation requires that the attacker already controls a spoke/participant cluster joined to the Submariner broker (reflected in CVSS PR:L), and that Submariner with the Lighthouse cross-cluster service-discovery component is deployed and actively syncing EndpointSlices/ServiceImports between clusters via the broker. The concrete trigger is the controller trusting an attacker-controllable label or annotation on the broker object to choose the destination namespace, so the attack only works where that resource-injection sync path is enabled across peers. Limiting factors: it is not exploitable from an unauthenticated external position - prior spoke-cluster compromise or rogue-participant status is a hard prerequisite - and clusters not using Submariner/Lighthouse multi-cluster service export are not exposed.

The signals are largely consistent and point to a genuine high-priority issue rather than an inflated CVSS. The vector CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H yields 9.9 driven by the scope change (S:C) - a compromised spoke reaching into peer clusters - combined with full C/I/A impact and low attack complexity. The important nuance is PR:L: this is not a from-zero remote exploit, it presupposes the attacker already controls a spoke cluster, so it is best understood as a blast-radius / lateral-movement amplifier that turns one compromised cluster into fleet-wide compromise. No EPSS score, KEV listing, or POC was provided, so there is no public exploit identified at time of analysis and no evidence of active exploitation; the low-complexity, high-impact nature nonetheless means any organization running affected ACM 2 with Submariner/Lighthouse should treat this as urgent patch-when-available. Absence of EPSS/KEV data is a data gap, not evidence of low risk - verify current KEV/EPSS status independently before deprioritizing.

An attacker first gains control of a single spoke cluster (via a separate compromise or a malicious/rogue participant), then sets a crafted label or annotation on the broker object so the destination namespace resolves to kube-system on a peer cluster. Submariner Lighthouse dutifully injects attacker-defined EndpointSlices and ServiceImports there, letting the attacker redirect or impersonate a system service and pivot to privilege escalation across the fleet. No public exploit code is identified at time of analysis, but the low attack complexity makes weaponization by a capable actor straightforward.