Vault CVE-2026-3605

Requires authenticated access to HashiCorp Vault with valid credentials (PR:L). The user's assigned policy must include at least one path rule for a KVv2 secrets engine mount that uses glob wildcard characters ('*' for multi-segment paths or '+' for single-segment paths) in the path definition, such as 'path "secret/data/*"' or 'path "kv/metadata/team-+/*"'. The vulnerability is specific to the KVv2 secrets engine-KVv1 and other Vault secrets engines are not affected. Attack is limited to secrets within the same Vault namespace as the attacker's policy scope; cross-namespace deletion is explicitly not possible per vendor confirmation. The misconfigured policy must grant some level of access (even minimal read) to the glob-matched path for the delete operation bypass to trigger. Default Vault installations without KVv2 enabled or deployments using only explicit path policies (no wildcards) are not vulnerable. Attack vector is network-based (AV:N) requiring only API connectivity to Vault servers, typically HTTPS on port 8200.

Real-world risk is moderate despite CVSS 8.1 score. The vulnerability requires pre-existing authenticated access (PR:L) with specific glob-based policy configurations, narrowing the attack surface compared to unauthenticated network exploits. EPSS score of 0.01% (2nd percentile) indicates minimal observed exploitation likelihood, aligning with SSVC assessment of 'none' for exploitation status and 'no' for automatable attacks. However, CVSS correctly reflects 'total' technical impact (SSVC) because successful exploitation enables unauthorized deletion of arbitrary secrets within a namespace, causing operational disruption. The attack complexity is low (AC:L) once conditions are met-no race conditions or cryptographic puzzles-but the prerequisite of valid credentials and glob-configured policies limits opportunistic mass scanning. Organizations using KVv2 with wildcard policies face tangible risk; those using explicit path grants or operating without KVv2 are unaffected. The 8.1 score appropriately weights insider threat scenarios where authenticated users might abuse policy loopholes. Patch availability from HashiCorp with multiple backported fixes reduces residual risk for maintained deployments, but the 6-year affected version span (0.10.0-1.x) suggests many unpatched instances exist in enterprise environments with slow upgrade cycles.

An insider threat actor or compromised service account with authenticated Vault access possesses a policy granting read access to 'secret/data/teamA/*' via glob pattern. The attacker crafts API calls to Vault's KVv2 delete endpoint targeting paths outside their read scope, such as 'secret/metadata/teamB/database-creds' or 'secret/data/teamC/api-keys'. Due to improper glob evaluation during delete authorization, Vault incorrectly permits these deletions despite the policy never granting write or delete permissions to teamB or teamC paths. The attacker systematically deletes critical secrets across multiple teams within the same namespace, causing application failures and operational outages when services attempt to retrieve non-existent credentials. The attack requires only standard Vault CLI or API knowledge, leaving no trace of exfiltrated secret values (confidentiality intact per CVSS C:N) but achieving high-impact denial-of-service across the organization's secrets infrastructure.