Fix library
Lock down your AI stack
Plain-English fixes for the most common AI-stack exposures. Each card: the risk, the fix in three steps, and a command to verify you’re done.
- CRITICAL Ollama — Ollama API open to the internet
- CRITICAL n8n — n8n instance open — settings or owner-setup reachable without login
- HIGH Open WebUI — Open WebUI reachable from the internet
- HIGH vLLM — vLLM API open to the internet
- MEDIUM Langfuse — Langfuse reachable from the internet
- CRITICAL ComfyUI — ComfyUI open to the internet
- CRITICAL Ray — Ray dashboard open to the internet
- CRITICAL n8n — n8n CVE-2026-21858 (Ni8mare) — unauthenticated file read / RCE
- CRITICAL Ollama — Ollama CVE-2024-37032 (Probllama) — remote code execution
- CRITICAL ComfyUI-Manager — ComfyUI-Manager CVE-2025-67303 — unauthenticated RCE
- CRITICAL Open WebUI — Open WebUI CVE-2026-44551 — LDAP empty-password auth bypass
- MEDIUM Langfuse — Langfuse CVE-2025-64504 — cross-organization user enumeration
- MEDIUM Langfuse — Langfuse CVE-2026-41487 — member role can steal LLM provider keys
- CRITICAL vLLM — vLLM CVE-2026-22778 — remote code execution via video endpoints
- MEDIUM vLLM — vLLM CVE-2026-54236 — memory addresses leaked in error messages
- CRITICAL Langflow — Langflow CVE-2026-33017 — unauthenticated RCE via public flow build