Exposed AI Infrastructure Surges 63% Amid Critical Industrial Risks

The report highlights that nearly 268,000 AI-related systems are now exposed globally. Workflow automation tools dominate this landscape, with n8n accounting for 114,000 of the identified hosts, followed by the personal AI agent assistant OpenClaw. Over half of this infrastructure is concentrated in the United States, China, and Germany, creating a dense target area for potential exploitation.

Beyond AI, the data points to a stubborn security gap in industrial infrastructure. Censys identified 134,000 exposed hosts running industrial control systems, with 70% residing on consumer or mobile networks, which complicates oversight for security teams. North America holds the largest share of these exposures at 38%, followed by Europe and Asia. While organizations are making progress in updating software like Microsoft SharePoint and Jenkins, the oldest, most vulnerable versions of these products remain stubbornly present on the open web.

Geographic patterns also define the current threat environment. While China hosts 36.6% of malicious infrastructure such as command-and-control servers, the United States remains the primary hub for web-based deception, hosting 45.5% of observed phishing and fraudulent pages. Zakir Durumeric, CEO of Censys, noted that identifying and continuously monitoring these Internet-facing systems is essential to balancing AI innovation with risk management.