{"id":"CVE-2026-41713","published":"2026-05-12T10:17:39Z","modified":"2026-05-12T12:11:27Z","title":"Prompt Injection via Memory Poisoning in PromptChatMemoryAdvisor","vendor":"VMware","product":"Spring AI","tier":"tail","description":{"text":"A malicious user could craft input that is stored in conversation memory and later interpreted by the model in an unintended way. Applications using the affected advisor with user-controlled input may be susceptible to manipulation of model behavior across conversation turns.","provenance":"source","from":"CVE.org"},"cvss":{"score":8.2,"severity":"HIGH","version":"3.1","vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:H/A:N","provenance":"source","from":"CVE.org CNA"},"epss":{"score":0.00218,"percentile":0.12513,"provenance":"source"},"cwes":[{"id":"CWE-1336","name":"Improper Neutralization of Special Elements Used in a Template Engine"}],"references":[{"url":"https://spring.io/security/cve-2026-41713","tags":[]},{"url":"https://nvd.nist.gov/vuln-metrics/cvss/v3-calculator?vector=AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:H/A:N&version=3.1","tags":[]}],"priority":{"score":8.2,"factors":[{"name":"EPSS percentile","points":2.5},{"name":"CVSS severity","points":5.74}]},"vendorSlug":"vmware","affectedVersions":[{"vendor":"VMware","product":"Spring AI","versions":[">= 1.0.0, < 1.0.7",">= 1.1.0, < 1.1.6"]}],"ssvc":{"exploitation":"none","automatable":"yes","technicalImpact":"partial","timestamp":"2026-05-12T12:11:07.556751Z","from":"CISA ADP","provenance":"source"},"affected":[{"ecosystem":"Maven","package":"org.springframework.ai:spring-ai-client-chat","vulnerableRanges":["<1.0.7",">=1.1.0-M1, <1.1.6"],"patched":["1.0.7","1.1.6"],"from":["GHSA-5852-phmh-8fhr"],"provenance":"source"}],"summary":{"text":"The flaw is a prompt injection weakness in the PromptChatMemoryAdvisor component of VMware's Spring AI product, where crafted user input can be stored in conversation memory. The stored content can later be interpreted by the model in unintended ways, so an attacker who can supply input to an application using this advisor with user-controlled data may be able to manipulate model behaviour across conversation turns, with no authentication or user interaction required over the network.","provenance":"ai","model":"claude-haiku-5-5","generatedAt":"2026-11-20T22:09:17Z"},"euvd":{"id":"EUVD-2026-29449","exploited":false,"provenance":"source"}}