Internet-exposed AI tools and industrial control systems (ICS) continue to grow in number, creating significant security vulnerabilities, according to a report by internet monitoring firm Censys. Hackers could exploit these internet-facing AI tools, which often contain serious vulnerabilities, to gain access to businesses. The expanding complexity of the AI ecosystem, where public services rely on a vast network of often neglected back-end products, introduces additional security risks.
The Censys report identified approximately 268,000 internet-connected devices running AI products. These products range from workflow automation tools to AI assistants and large language model gateways. The vast majority, 82%, of these devices are running five popular products, including the workflow automation platform n8n and the chatbot front-end interface Open WebUI. This highlights the increasing complexity of the AI environment, a major factor hindering security efforts.
AI Tools Present Growing Attack Surface
Censys analysts noted an emerging characteristic AI infrastructure stack on the internet. This stack includes not only exposed LLM interfaces but also supporting infrastructure layers for building and operating AI applications. Each widely exposed AI tool carries unique risks. For instance, a significant vulnerability in OpenClaw could allow arbitrary code execution on a host machine. N8n, an AI orchestration tool, also experienced a notable vulnerability year, including a serious arbitrary file-read vulnerability.
The attack surface expands not just with new AI applications, but also when each supporting layer becomes independently reachable from the internet, requiring dedicated security measures. Between October 2025 and June 2026, the number of internet-exposed AI tools surged by 63%, peaking in early 2026 with nearly 300,000 endpoints. The United States had the largest share of these exposures, with 62,687 IP addresses, followed by China and Germany. Cloud service providers accounted for a significant portion of exposures, with approximately 32,000 appearing on Amazon’s infrastructure alone, and the top five cloud providers sharing a similar amount of exposure.
Industrial Control Systems Also Vulnerable
The Censys report also revealed that a large number of ICS devices are accessible from the internet. Censys scans identified approximately 134,000 ICS devices in early 2026, a 4% increase from 2024 figures. More than 80% of these devices run non-ICS functions, such as HTTP and FTP servers, increasing intrusion risks. Hackers may bypass complex ICS protocols by breaching targets through simpler IT protocols.
A concerning finding is that roughly 70% of exposed ICS devices operate on mobile or consumer networks, complicating identification and warnings to operators. Industrial devices using cellular modems are often in remote locations, potentially lacking frequent security updates. This increased attention to digital vulnerabilities in critical infrastructure follows recent Iran-linked hacks of water systems and China-linked campaigns against various sectors.
In some countries, a single telecommunications company hosts the majority of exposed ICS devices, creating a significant concentration risk. For example, in Turkey, 81% of internet-accessible devices use one mobile provider. Censys stated that such homogeneous hosting patterns can simplify reconnaissance and hunting for threat actors.
In the U.S., exposure is more dispersed, with Verizon hosting one-third of internet-accessible devices. North America accounts for about 38% of all identified exposed ICS devices, followed by Europe at 32% and Asia at 25%. The complexity of securing these interconnected systems, particularly when traditional IT services are integrated into ICS, presents ongoing challenges for cybersecurity professionals. The trend of increasing exposure suggests a need for enhanced monitoring and security protocols for both AI tools and critical infrastructure.
