LLMVault is a deliberately-insecure training range for the OWASP Top 10 for LLM Applications. The vulnerabilities in the labs are by design and are not security issues to report. Please do not file reports about the labs being exploitable — that is the entire point.
- Do not deploy LLMVault on the public internet. Run it locally or on an isolated,
access-controlled host. Bind to
127.0.0.1where possible. - The expert tier ships encrypted; its flags and the access key are never in this repository. Do not attempt to obtain the key by means other than earning it.
- Treat any credentials/keys you use with the optional Ollama integration as local-only.
Please open a GitHub issue (or a private security advisory) if you find:
- An accidental secret leak in the repository (a real key, solution, or the operator vault committed by mistake).
- A flaw in the harness itself that lets the expert tier decrypt without the key, or that leaks progress/data beyond the local instance.
- A supply-chain problem in the project's own dependencies or CI.
The latest main is the supported version.