Munim releases vulnerability checking tool for self-hosted AI stacks, covering security advisories missing from public databases such as OSV.dev
Self-hosted AI Stack Vulnerability Check
Munim, Inc. released a vulnerability checking tool for self-hosted AI stacks, with the research and code completed by an AI agent; input the versions of applications such as Ollama, vLLM, n8n and Dify or packages such as PyPI and npm, and it returns which known advisories affect each component, their severity and the versions to upgrade to. Most scanners only read OSV.dev and the GitHub Advisory Database, but many advisories for such applications have no CVE ID or are distributed as Docker images and do not make it into these databases; the tool merges manually curated advisory data with real-time OSV.dev queries, for example vLLM 0.30.0 is affected by remote code execution when trust_remote_code is enabled (CVSS 8.1, fixed in 0.31.0). The tool checks at most 500 components per run and also offers a GitHub Action to check versions pinned in a repository; it does not connect to servers or scan networks, and versions with no known advisories may still be insecure when exposed without authentication.
The tool compiles security advisories for self-hosted AI applications not included in public databases such as OSV.dev, which teams building their own AI stacks can use to check versions and upgrade.