RabbitMQ CVE-2026-67233

Requires BOTH the rabbitmq_shovel and rabbitmq_shovel_management plugins to be enabled (a non-default configuration), and the attacker must already hold valid credentials carrying the monitoring tag - intended as read-only. Given those, exploitation is a single authenticated network DELETE request against the shovel management resource with no user interaction; it is limited to dynamic shovels in vhosts the monitoring user can see. Limiting factors: the plugin dependency, the need for pre-existing (even if low-privilege) credentials, and visibility scoping to permitted vhosts.

This is a legitimate but bounded broken-authorization flaw (CWE-862), not a high-CVSS-but-empty issue. The CVSS 4.0 vector (AV:N/AC:L/AT:P/PR:L/UI:N/VC:N/VI:H/VA:L, 6.0 MEDIUM) aligns well with the description: a network-reachable, low-complexity operation requiring already-provisioned monitoring credentials (PR:L) and a specific plugin precondition (AT:P). Impact is integrity-primary (VI:H - arbitrary shovel deletion/restart) with limited availability (VA:L - disrupted replication) and zero confidentiality (VC:N), which the vector captures accurately. There is no EPSS score, no CISA KEV listing, and no public exploit code identified at time of analysis, so exploitation is neither confirmed nor probable in the wild - the risk is that of a privilege-boundary defect exploitable only by an existing tenant who was deliberately granted read-only monitoring access. The AT:P (attack requirement) is the key limiter: both rabbitmq_shovel and rabbitmq_shovel_management plugins must be enabled, which is a non-default deployment. No signals conflict; all point to a real-but-scoped insider/limited-privilege risk rather than an internet-wide emergency.

An organization grants a monitoring/dashboard user read-only RabbitMQ credentials (monitoring tag) for observability. That user - or anyone who compromises those low-value credentials - issues a DELETE against the shovel management API and tears down or restarts dynamic shovels in any vhost they can see, silently breaking cross-cluster/federated message replication. No public exploit code is known, but the request is a trivial authenticated HTTP DELETE requiring no special tooling.