Vulnerability  ·  2026-08-28

Spring AI — PDF ingest stack overflow and predictable ONNX model-cache path substitution

VulnerabilityMedium impactGlobal
Spring AI (2.0.0, 1.1.0–1.1.8, 1.0.0–1.0.9) contains CVE-2026-47851, where analyzing a PDF with a deeply nested or cyclic table-of-contents outline tree causes unbounded recursion and a StackOverflowError in the ingestion thread, and CVE-2026-47852, where a local attacker on a multi-user host can pre-create Spring AI's deterministic ONNX model cache path and plant a malicious model file that gets loaded by other users/processes.
Spring AI is a widely used Java framework for building RAG/document-ingestion pipelines; the PDF vulnerability enables a trivial denial-of-service against document-ingestion services, while the ONNX cache-path issue lets a local attacker on shared infrastructure substitute a malicious embedding model that other tenants unknowingly load and execute.
Uploading a crafted PDF with cyclic/deeply nested TOC to trigger stack overflow; or pre-creating a predictable cache file path to plant a malicious ONNX model on a shared host
Spring AI 1.0.0-1.0.9, 1.1.0-1.1.8, 2.0.0
Upgrade to Spring AI 2.0.1 or later per spring.io/security/cve-2026-47851 and cve-2026-47852
Spring AI 2.0.1 release notesNVD - CVE-2026-47851NVD - CVE-2026-47852
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →