1M Open AI Services: How Vulnerable Is AI Infrastructure

Colleagues, a cybersecurity alert: Intruder scanned over 1 million publicly accessible AI services and uncovered widespread vulnerabilities.

Many deployments lack authentication — data and tooling are externally exposed.
Open chatbots and orchestration platforms (Flowise, n8n) leaked prompts, logs and credentials; 90+ instances found in government, marketing and finance.
Of ~5,200 Ollama API endpoints, 31% responded; 518 servers hosted known frontier models.
Common issues: hard‑coded credentials, services running as root, insecure Docker configurations and RCE vectors.

Why it matters: these flaws enable data exfiltration and full infrastructure compromise.
Have you assessed your AI services for these risks?

#cybersecurity #AI #APIsecurity #infrastructure