Censys Report Finds Exposed AI Infrastructure Grew 63% in Nine Months
Exposed AI infrastructure grew by 63% in nine months, reaching nearly 268,000 internet-facing systems, according to the fifth annual State of the Internet Report released by Censys. The report also identified approximately 134,000 internet-exposed hosts running industrial control systems (ICS), around 70% of which were found on consumer and mobile networks.
The findings highlight the expanding attack surface associated with AI adoption, alongside persistent industrial technology exposures and uneven progress in software maintenance. The report also identifies geographical differences in the distribution of internet-connected infrastructure associated with malicious activity.
The Censys Advanced Research Collective (ARC) analysed historical and real-time data from the Censys Internet Map, which continuously monitors global internet infrastructure, to assess changes in exposed systems, emerging technologies and threat infrastructure.
AI infrastructure expands rapidly
The report found that internet-facing AI tools increased by 63% over nine months, reaching nearly 268,000 systems. These include not only chatbots but also tools connecting AI applications with data, business applications and user accounts.
Workflow automation platform n8n was the most widely exposed AI-related product, accounting for nearly 114,000 hosts. OpenClaw, a personal AI agent assistant, followed with approximately 54,000 hosts.
The United States accounted for 62,687 exposed AI systems, followed by China with 42,511 and Germany with 31,222. Together, the three countries represented just over half of the exposed AI infrastructure identified by Censys.
Industrial control systems remain exposed
The report identified approximately 134,000 internet-exposed hosts running ICS devices and services. North America accounted for around 38% of the global exposed ICS landscape, followed by Europe at 32% and Asia at 25%.
Approximately 70% of these exposed hosts were consistently found on consumer and mobile networks, potentially complicating efforts to identify system owners and address vulnerabilities.
The findings highlight the challenges of securing industrial environments as operational technology remains accessible through internet-connected infrastructure.
Threat infrastructure varies by geography
Censys also identified differences in the geographical distribution of malicious infrastructure. China accounted for 36.6% of observed host-level malicious infrastructure, including command-and-control systems and remote access tools, compared with 16.8% in the United States.
The distribution was different for web-based threats, including phishing and deceptive websites. The United States hosted 45.5% of the web-based threats observed by Censys, while China accounted for 20.9%.
Software maintenance presented another concern. The report examined Progress Telerik Report Server, Jenkins, Cleo MFT and Microsoft SharePoint Server, finding that organisations had adopted newer versions of the products, but the oldest versions remained largely unchanged throughout the study period.
Zakir Durumeric, Co-Founder and CEO of Censys, said, “The Internet is constantly changing, and understanding how it changes is critical to modern cybersecurity. New technologies such as AI introduce new layers of infrastructure. Identifying and continuously monitoring Internet-facing systems is essential to realizing AI’s value without introducing unmanaged risk.”
The report underscores the importance of maintaining visibility into internet-facing assets as organisations adopt AI technologies and manage existing infrastructure. Continuous monitoring can help security teams identify exposed systems, assess emerging risks and prioritise remediation across increasingly complex environments.
