CVE-2026-59344

Spring AI Coherence filters could bypass tenant or document scope

Spring / Broadcom · Spring AI · CoherenceVectorStore

First documented disclosure · Review checkpoint

Summary

Crafted filter keys could return documents outside intended restrictions.

Severity and exploitation evidence

High · 7.5/10 · CVSS 3.1. Severity source.

Exploitation not established: The vendor describes the flaw and remediation; the reviewed sources do not establish malicious exploitation.

Exploitation evidence source

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

Quoted filter keys are passed as Coherence extractors rather than validated literal metadata names.

Conditions required

Externally influenced or model-generated filter expressions reach a Coherence-backed vector store.

Potential impact

Disclosure of documents outside the filter’s intended scope.

Affected versions and configurations

2.0.0–2.0.2, 1.1.0–1.1.9 and 1.0.0–1.0.10.

Fix and mitigation status

2.0.3 (OSS); 2.0.2.1 (enterprise support only). Older supported branches: 1.1.10 and 1.0.11, enterprise support only.

Practical next steps

Upgrade to the applicable corrected release. Public 2.0.3 is available; listed older-branch binaries require enterprise support.

Operating lesson

Apply explicit bounds and trust checks to every path accepting untrusted AI context or input.

Evidence and limits

Original vendor advisory is November 19. All nine exact CVE authority requests returned 404 at review; the numbering and affected/fixed mapping come directly from Spring. GitHub 2.0.3 release was published November 20 at 15: 47: 36Z, before this review cutoff. CVSS 3.1 base scores are calculated from each Spring advisory’s published vector; the vendor severity label is retained separately. No score is borrowed from another source.

Sources