OpenStack Zaqar CVE-2026-107363

Exploitation requires that the deployment uses the Zaqar WebSocket transport together with Keystone authentication - the REST transport and non-Keystone setups are explicitly out of scope. The attacker must already hold a valid Keystone token for at least one project (PR:L, authenticated) and must know or be able to enumerate the target project's UUID (reflected as AT:P). No end-user interaction is needed (UI:N). Limiting factors that reduce real-world exposure: the WebSocket transport must be deliberately enabled (not the most common Zaqar configuration), the attacker needs legitimate tenant credentials first, and value is highest only in multi-tenant clouds where crossing the project boundary yields access to other tenants' queue data.

This is a moderate-priority authorization flaw, not a top-tier emergency. The vendor CVSS 4.0 vector (AV:N/AC:L/AT:P/PR:L/UI:N/VC:L/VI:H/VA:H) scores 6.1 (MEDIUM) and the signals are internally consistent: network-reachable but NOT unauthenticated - PR:L confirms an attacker needs a valid Keystone token for at least one project, and AT:P reflects an additional attack requirement (knowledge of a target project's UUID). SSVC reinforces a measured posture: Exploitation=poc (proof-of-concept only, not active exploitation), Automatable=no (each cross-tenant target requires a known project UUID and cannot be sprayed at internet scale), and Technical Impact=partial. There is NO CISA KEV listing and NO EPSS score was provided, so widespread opportunistic exploitation is not indicated. The real risk is highest in multi-tenant OpenStack clouds where WebSocket transport is enabled - a malicious or compromised low-privilege tenant can cross the project isolation boundary, which is a serious trust-model violation despite the medium score. Conflict to verify: the description's primary harm is cross-tenant queue create/delete (integrity/availability), consistent with VI:H/VA:H, while confidentiality is rated only VC:L even though queue 'enumerate/inspect' implies some data disclosure - defenders handling sensitive queue payloads should treat confidentiality impact as potentially higher than the vector suggests.

A legitimate low-privilege tenant (or an attacker who has phished one tenant's Keystone credentials) connects to the Zaqar WebSocket endpoint with their valid token, then supplies a different project's UUID in subsequent frames. Because the transport never re-binds the request to the token's authenticated project, the attacker enumerates, reads, creates, and deletes queues belonging to an unrelated tenant. A proof-of-concept is reported (SSVC Exploitation=poc), and the moderate complexity stems only from needing to know or discover a target project UUID.