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

WeaknessASC-2026-0015Major weakness

Exposed AI infrastructure without authentication

CriticalFixed

Inference servers, agent builders, notebooks and cluster dashboards (Ray, Ollama, vLLM, Langflow, n8n, Jupyter) often ship without authentication. Internet scanners find them within hours and attackers run code or steal the keys they hold.

How to detect it

  • An AI service port (Ray 8265, Ollama 11434, Langflow 7860, vLLM 8000, Jupyter 8888, n8n 5678) answers from outside your network
  • The service accepts API calls or opens its UI without a login
  • Versions with known exploited CVEs (Langflow CVE-2025-3248, Ray CVE-2023-48022)

How to resolve it

  • Bind AI services to localhost or a private network and reach them through an authenticating gateway or VPN
  • Firewall dashboard and job APIs; never expose them publicly
  • Patch to versions without known exploited CVEs and re-check on every advisory
  • Scan your own external address space for AI ports on a schedule

How to investigate

  1. Check internet scan engines (Shodan, Censys) and your own scans for your addresses on AI ports
  2. Review access logs and job history for unknown sources, miners or new workloads
  3. Rotate every API key and credential the service stored

Sources

Open it in the Learning Center's vulnerability database