We Scanned 1 Million Exposed AI Services. Here's How Bad the Security Actually Is

Intruder's investigation of self-hosted AI infrastructure found widespread insecure defaults and misconfigurations—many instances lacked authentication, exposed chat histories and API keys, and revealed accessible agent management platforms (e.g., Flowise, n8n) and unsecured Ollama APIs. The scan and lab analysis showed significant scale and impact potential (exposed tooling, credentials, and at least one arbitrary code execution finding), highlighting that rapid AI adoption has outpaced security best practices and leaving organizations at high risk of data exposure and compromise.