CVE-2026-9741: MongoDB Queryable Encryption & CSFLE Plaintext Field Leak in Vector Search

Technical summary

CVE-2026-9741 is a cleartext transmission vulnerability (CWE-319) affecting MongoDB's query analysis layer during $vectorSearch aggregation processing. When a $vectorSearch stage includes filter expressions containing literal values from fields marked for Queryable Encryption or CSFLE protection, the query analyzer fails to preserve the encryption boundary and transmits these literals as plaintext to the server instead of as ciphertext. The vulnerability manifests only in queries that combine vector search with encrypted field filters; standard encrypted queries without vector search are unaffected. The root cause lies in incomplete integration of encryption handling into the vector search code path during query compilation and validation.

Business impact

Organizations relying on MongoDB encryption to meet data protection compliance requirements (HIPAA, PCI-DSS, GDPR) face unexpected exposure of encrypted field values in vector search queries. This could lead to compliance violations if unencrypted sensitive data (credit cards, PHI, PII) transits the network or appears in server logs. The impact is limited to authenticated users who both (1) have permission to execute $vectorSearch queries and (2) target encrypted fields in filter expressions. However, insider threat risk increases: a DBA or compromised application account could exfiltrate encrypted field values by crafting specific vector search queries.

Affected systems

MongoDB products using Queryable Encryption or Client-Side Field Level Encryption are affected. The vulnerability applies to any version that implements $vectorSearch with QE/CSFLE; consult MongoDB's official security advisory for precise version ranges and patched builds. Applications using CSFLE or QE for sensitive field protection but not using $vectorSearch aggregation are unaffected. Atlas Vector Search users should verify their encryption configuration and query patterns.

Exploitability

The vulnerability requires authentication: an attacker must have a valid database connection and permission to run aggregation queries. No network exploit exists; the issue is triggered by intentional query construction combining $vectorSearch with encrypted field filters. Exploitation does not require special tools—standard MongoDB client libraries and query syntax suffice. However, the attacker must understand which fields are encrypted and how to reference them in filter expressions. Public exploit code is unlikely because successful exfiltration depends on application-specific field names and encryption configuration. CVSS 3.1 score of 6.5 (Medium) reflects the authentication requirement and high confidentiality impact balanced against limited availability/integrity effects.

Remediation

Apply MongoDB security patches as released by the vendor. MongoDB will issue patched versions that correctly encrypt literal values in $vectorSearch filter expressions. Interim mitigations include: restrict $vectorSearch query execution to trusted application code paths; audit and monitor aggregation queries that combine $vectorSearch with encrypted fields; consider disabling $vectorSearch if vector search functionality is not required. Verify patch applicability and compatibility before production deployment.

Patch guidance

Monitor MongoDB's security advisories and release notes for patched versions addressing CVE-2026-9741. Patches will correct the query analyzer to preserve encryption semantics when processing $vectorSearch filter expressions. Verify patch version numbers and compatibility with your MongoDB deployment model (self-managed or Atlas) before applying. Test patched versions in a staging environment to ensure no regression in vector search functionality or performance. Rolling updates may be required for replica sets and sharded clusters to maintain availability.

Detection guidance

Search application logs and database audit logs for aggregation queries containing both $vectorSearch and filter conditions on fields marked for QE/CSFLE encryption. Monitor network traffic for plaintext transmission of values that should be encrypted according to your encryption policy. Enable MongoDB query logging and filter for aggregation stages containing encrypted field references. Correlate logs with database schema metadata to identify queries that violate encryption boundaries. Intrusion detection rules should flag repeated $vectorSearch queries targeting the same encrypted fields, which may indicate reconnaissance or data exfiltration attempts.

Why prioritize this

Patch this vulnerability as part of your regular security update cycle (within 60–90 days), prioritizing test and staging environments first. While CVSS 6.5 is Medium severity, the confidentiality impact is high for organizations that depend on encryption to protect sensitive fields in vector search use cases. If your applications do not use $vectorSearch with encrypted fields, the immediate risk is lower but patches should still be applied to maintain defense in depth. Prioritize patching if you operate in regulated industries (healthcare, finance) where unexpected plaintext transmission of sensitive data could trigger compliance reviews or incident notifications.

Risk score, explained

CVSS 3.1 score of 6.5 (Medium) is driven by: high confidentiality impact (encrypted field values leak as plaintext), no integrity or availability impact, and authentication requirement (PR:L). The score appropriately reflects that while the vulnerability is serious for data protection, it requires a valid database user to exploit and does not cascade to system compromise. The lack of CISA KEV designation indicates no evidence of active exploitation in the wild as of the publication date, reducing immediate tactical urgency but not strategic importance for organizations with relevant workloads.

Frequently asked questions