CVE-2026-40966

VectorStoreChatMemoryAdvisor conversation scoping can lead to cross-tenant memory exfiltration

Published Apr 28, 2026 CVSS 3.1 · 5.9 Medium EPSS < 1% KEV · Not listed
Description
In Spring AI, an attacker can bypass conversation isolation and exfiltrate sensitive memory from other users’ chat histories, including secrets and credentials, by injecting filter logic through conversationId. Only applications that use VectorStoreChatMemoryAdvisor and pass user-supplied input as a conversationId are affected.

Analysis

AI-generated

Technical impact

VectorStoreChatMemoryAdvisor in VMware Spring AI allows an attacker to inject malicious filter logic through the conversationId parameter, bypassing conversation isolation and extracting chat history, including secrets and credentials, from other users. This flaw is an example of improper access control, enabling cross‑tenant data leakage without authentication.

Affected Systems

The vulnerability affects systems that deploy VMware Spring AI with the VectorStoreChatMemoryAdvisor component and accept user‑provided conversation identifiers. No specific version numbers are listed; any application that relies on this advisor to scope chat memories is potentially impacted.

Risk and Exploitability

The CVSS score of 5.9 indicates moderate severity, and the absence of an EPSS rating does not provide an estimated exploitation probability. The vulnerability is not currently catalogued in CISA’s KEV list. Based on the description, the likely attack vector is via a network‑based API where an attacker supplies a crafted conversationId to manipulate the filter logic. Successful exploitation could lead to unauthorized disclosure of sensitive user data.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Apply the latest VMware Spring AI patch or update to a version where the VectorStoreChatMemoryAdvisor enforces proper isolation of conversation data.
  • Sanitize and validate all user‑supplied conversationId values to prevent injection of filter expressions that could bypass tenant boundaries.
  • Enforce strict access controls so that only authenticated, authorized users can access chat histories for their own tenant, limiting the scope of data the advisor can retrieve.

Generated by OpenCVE AI on April 28, 2026 at 12:24 UTC.

Tracking

Sign in to view the affected projects.

Advisories

Source ID Title
Github GHSA Github GHSA GHSA-v6x6-pjxw-3pv2 Spring AI's VectorStoreChatMemoryAdvisor conversation scoping can lead to cross-tenant memory exfiltration
History
Type Values Removed Values Added
CPEs cpe:2.3:a:vmware:spring_ai:*:*:*:*:*:*:*:*
Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'no', 'Exploitation': 'none', 'Technical Impact': 'partial'}, 'version': '2.0.3'}

Type Values Removed Values Added
First Time appeared Vmware
Vmware spring Ai
Vendors & Products Vmware
Vmware spring Ai
Type Values Removed Values Added
Description In Spring AI, an attacker can bypass conversation isolation and exfiltrate sensitive memory from other users’ chat histories, including secrets and credentials, by injecting filter logic through conversationId. Only applications that use VectorStoreChatMemoryAdvisor and pass user-supplied input as a conversationId are affected.
Title VectorStoreChatMemoryAdvisor conversation scoping can lead to cross-tenant memory exfiltration
Weaknesses CWE-284
References
Metrics cvssV3_1

{'score': 5.9, 'vector': 'CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:N/A:N'}

Weaknesses

Sources

Status
PUBLISHED
Assigner
vmware
Published
Updated
Apr 28, 2026, 1:37 PM
Reserved
Apr 16, 2026, 2:18 AM
Status
Analyzed
Published
Apr 28, 2026, 8:16 AM
Modified
Oct 26, 2026, 9:10 PM