Executive brief
Spring AI, a framework for building AI-powered applications, contains a vulnerability that could allow an attacker to view the private chat histories of other users. By providing a specially crafted conversation ID, an attacker can bypass security boundaries and access sensitive information, such as secrets or credentials, stored in the application's memory. This risk primarily affects multi-user applications that use the VectorStoreChatMemoryAdvisor component and allow users to specify their own conversation identifiers.
Technical details
A filter expression injection vulnerability exists in Spring AI's VectorStoreChatMemoryAdvisor. The root cause is the improper sanitization of the 'conversationId' parameter, which is concatenated into filter logic without adequate escaping. An attacker can exploit this by providing a malicious conversationId that alters the filter logic, bypassing intended isolation to query and exfiltrate chat records belonging to other users or tenants. This vulnerability requires the application to be configured to use VectorStoreChatMemoryAdvisor and to accept user-supplied input for the conversationId. Patches are available in versions 1.0.6 and 1.1.5.
Affected products
- Spring spring-ai-advisors-vector-store >= 1.0.0, < 1.0.6; >= 1.1.0, < 1.1.5
Timeline
- 2026-04-28: disclosed
- 2026-04-28: advisory
- 2026-05-06: other: GitHub advisory reviewed