Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

Best practiceModels and infrastructure

Never expose AI infrastructure to the internet

OWASP ASI05OWASP ASI03

Ray, Ollama, vLLM, Langflow, n8n, MCP inspectors and vector databases were built for trusted networks and often ship without authentication. Exposed instances are found by scanners within hours and turned into botnets, crypto miners and credential harvesters.

What to do

  1. Put every AI service behind a VPN, private network or an authenticating proxy; bind local tools to 127.0.0.1.
  2. Assume 'no auth by default' until you have read the docs; Ray, Ollama and MCP inspectors have all shipped that way.
  3. Scan your own external attack surface for AI service ports and dashboards, and alert on new ones.
  4. Apply the same rule to developer laptops: an inspector or a local MCP server listening on 0.0.0.0 is exposed on café Wi-Fi.

Sources

Check your own AI plan against the practices: Plan check