Get ahead
VMware offers training and certification to turbo-charge your progress.
Learn moreApplications using spring-ai-redis-semantic-cache with a SemanticCacheAdvisor shared by more than one end user are vulnerable to cross-user information disclosure and cache poisoning. Cached responses were isolated only by the hash of the system prompt, with no user, session, or tenant dimension, so all users shared a single cache pool.
A cached response that contains one user's data can be served to any other user who asks a similar question. An attacker can also prime a cache entry with attacker-controlled content, which is then served to later users. Cached entries do not expire when written through the advisor.
This issue is a residual isolation gap related to CVE-2026-59308, which addressed only isolation by system prompt.
Spring AI:
| Fix version | Availability |
|---|---|
| 2.0.3 | OSS |
| 2.0.2.1 | Enterprise Support Only |
| 1.1.10 | Enterprise Support Only |
| 1.0.11 | Enterprise Support Only |
Users of affected versions should upgrade to the corresponding fixed version.
The fixed versions introduce a mandatory SemanticCacheScopeResolver that determines the scope in which cached responses are shared. Applications must define one that derives the scope from a trusted source, such as the authenticated principal, and not from a value the end user can choose.
If upgrading is not immediately possible, do not use SemanticCacheAdvisor in applications that serve more than one end user or tenant.
This vulnerability was responsibly reported by Sai Sreewathsa Kovalluri and Kannan K (Researchers) of Innspark Solutions (https://innspark.in) and independently reported by Daehyun Kang @daehyuh
To report a security vulnerability for a project within the Spring portfolio, see the Security Policy