CVE-2026-125327
Query processing could exhaust resources by rebuilding the keyword index
pgEdge · RAG Server hybrid search
First documented disclosure · Review checkpoint
Summary
The search worker loaded an unbounded corpus and rebuilt its in-memory index per request. Database, CPU and memory demand could make retrieval unavailable.
Severity and exploitation evidence
High · 7.5/10 · CVSS 3.1. Severity source.
Exploitation not established: The reviewed advisory does not report malicious exploitation.
Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.
How the weakness works
The search worker loaded an unbounded corpus and rebuilt its in-memory index per request.
Conditions required
The unauthenticated query API was reachable with hybrid search, which was enabled by default.
Potential impact
Database, CPU and memory demand could make retrieval unavailable.
Affected versions and configurations
Maintainer lists releases through 1.0.0. CNA describes versions before 2.0.0.
Fix and mitigation status
2.0.0.
Practical next steps
Upgrade RAG Server and restrict or rate-limit public access to retrieval queries.
Operating lesson
Give AI retrieval and tool services database credentials scoped to the data they need.
Evidence and limits
The public disclosure was on 2026-09-03. The CVE record was published on 2026-11-10.
Sources
- pgEdge maintainer advisory
Primary source ·
- GitHub_M CVE record
Primary source ·