GHSA-2mfg-cc43-9pcj: LangChain4j: SQL injection via metadata filters in langchain4j-mariadb and langchain4j-pgvector

Summary

The langchain4j-mariadb and langchain4j-pgvector embedding stores build metadata-filter SQL by concatenating filter keys, and in MariaDB string values, without adequate escaping. A crafted key in EmbeddingSearchRequest.filter() can inject arbitrary SQL into search() and removeAll(Filter) operations. Applications that let attacker-influenced filter keys reach these stores, such as LLM-generated filters, are exposed to blind data exfiltration, denial of service and deletion of arbitrary rows. Applications using only hard-coded filter keys are not reachable.

Mitigation

Fixed in langchain4j-mariadb and langchain4j-pgvector 1.16.3-beta26. Workarounds: do not pass untrusted input as metadata filter keys, and restrict filter keys to a known allow-list at the application layer.

Related items

• MediumCVE-2026-121936: easyVoice server-side request forgery through openaiBaseUrl in /api/v1/ttsSimilar attack · NVD/CVE Database
• MediumCVE-2026-118628: MLflow get-artifacts server-side request forgery information disclosureSimilar attack · NVD/CVE Database
• HighCVE-2026-118625: MLflow Tracking Server privilege escalation via get_endpointsSimilar attack · NVD/CVE Database
• InfoWhen trust becomes the payload: Inside the threats, etc quarterly reportSimilar attack · Elastic Security Labs
• MediumCritical Langflow flaws enable authentication bypass and RCESimilar attack · CSO Online