Temporal Server CVE-2026-104355

Requires a multi-cluster Temporal Server deployment in which an administrator has registered at least one remote cluster with connection enabled, AND the attacker must operate or compromise such a registered peer so it can control the replication replies the receiving cluster pulls. Given that position, NO credential on the receiving cluster is needed: the attacker only needs to supply a replication task whose namespace name and identifier match an existing local namespace (including the built-in temporal-system namespace). Limiting factors: a single-cluster deployment, or any cluster with no remote cluster registered with connection enabled, is explicitly not exposed; exploitation targets only local (non-global) namespaces; and the attacker must already hold the trusted peer-cluster position, which is the dominant real-world constraint.

This is a genuine but conditionally-scoped authorization flaw (CWE-863), not an inflated-CVSS paper risk. The vendor-supplied CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:H/UI:N, VI:H/VA:L) scores 7.0 HIGH and is internally consistent with the description: network-reachable replication path, low complexity, but requiring a high-privilege position (PR:H) - control of a cluster the victim's administrator has deliberately registered as a connected peer. The integrity impact is real and significant (VI:H): an attacker can silently overwrite any local namespace - including the built-in temporal-system namespace - changing its state, owner, retention, archival, cluster list, and active cluster, after which new Workflow starts fail (hence VA:L availability impact). There is NO confidentiality impact (VC:N). No EPSS score, KEV listing, or POC is provided in the input, so exploitation status is 'no public exploit identified at time of analysis' - this is a trust-boundary design bug, not an observed in-the-wild campaign. The decisive mitigating signal is the PR:H precondition plus the deployment gate: a cluster with no remote cluster registered with connection enabled is explicitly NOT exposed. This narrows the realistic population to multi-cluster/global-namespace Temporal deployments where a peer is malicious or compromised - a trusted-insider or supply-chain-of-trust scenario rather than mass opportunistic exploitation. Net: prioritize patching for anyone running connected multi-cluster Temporal, but single-cluster operators are unaffected. Note the input Tag 'Authentication Bypass' is slightly imprecise - no credential on the receiving cluster is bypassed; rather the receiving side fails to AUTHORIZE the scope of an otherwise-trusted peer's replication task (CWE-863, improper authorization).

An attacker who operates or has compromised a Temporal cluster that the victim's administrator registered as a connected peer crafts an inbound namespace replication task whose name and identifier match an existing LOCAL namespace (e.g. temporal-system) on the receiving cluster. Because the receiving side never checks that the namespace is global, it overwrites that namespace's state, retention, owner, and active cluster - marking it Deprecated or re-pointing its active cluster - so subsequent Workflow starts in it fail. No public exploit is identified, though the vulnerable logic and fix are public via the vendor patch PRs.