libmongocrypt CVE-2026-84962

The attacker must have write access to the MongoDB key vault collection, which is a highly privileged database-level position (PR:H in CVSS 4.0). The target deployment must have Client-Side Field Level Encryption (CSFLE) or Queryable Encryption explicitly enabled and configured with a Google Cloud KMS key provider (kmsProviders: { gcp: { ... } }); default MongoDB deployments without this feature are not affected. A specific prerequisite condition must exist (AT:P in CVSS 4.0), likely meaning the malicious key vault document must be in place before the authorized client performs a KMS-touching operation. An authorized client application must then actively perform an encrypt or decrypt operation that causes libmongocrypt to read the malicious key vault entry and construct an outbound KMS API request (UI:P). The following factors limit exploitation scope: write access to the key vault is a privileged position not typically granted to end-user accounts; exploitation requires an active cryptographic operation by the victim client to occur while malicious data is present; and only GCP KMS-backed CSFLE/Queryable Encryption deployments are in scope.

The CVSS 4.0 vector (AV:N/AC:H/AT:P/PR:H/UI:P, score 5.7) accurately models this as a moderate-severity, multi-prerequisite attack: the adversary must hold key vault write access (PR:H), a specific system prerequisite must be satisfied (AT:P), and the authorized client must actively trigger a cryptographic operation while malicious data is in place (UI:P). The raw score understates business impact for affected deployments - VC:H represents access to the cloud key management plane, which entirely defeats the security guarantee of client-side encryption; any organization that deployed CSFLE specifically to ensure the database operator cannot read plaintext data loses that protection entirely. The attack is insidious because it proxies through a legitimate, credentialed client process, making it difficult to distinguish from normal KMS traffic in audit logs. Even a mid-sized municipal IT team in inowroclaw relying on MongoDB CSFLE to protect sensitive records should treat remediation as urgent if any service account or administrator holding key vault write access is potentially compromised. No KEV listing and no public exploit code are confirmed, but the high confidentiality impact on key material and the silent, proxy-based nature of exploitation elevate practical risk well above what the 5.7 base score suggests for in-scope deployments.

An attacker with MongoDB database administrator access (or a compromised service account with key vault write permissions) modifies a key vault document to embed CRLF sequences in a KMS metadata field such as the key resource name or endpoint URI. When a legitimate application client subsequently performs a CSFLE decryption or encryption operation, libmongocrypt constructs a Google Cloud KMS HTTPS request incorporating the injected content, resulting in an arbitrary KMS API call - such as creating or destroying a key or decrypting an unrelated ciphertext - authenticated with the application's GCP service account credentials and invisible to standard application-level auditing.