AI‑Related Software Exposure Grows 63% in Nine Months, Accelerating the Internet Attack Surface
What Happened — Censys’ internet‑wide scans show AI‑related services exposed to the public internet jumped from 164,088 IP addresses in Oct 2025 to 267,915 in Jul 2026 – a 63 % increase. The surge is driven by fast‑moving products such as n8n (over 113k hosts, 76 % running versions with critical ≥9.0 CVEs) and the newly released OpenClaw AI agent, which went from 2 to 85,478 exposed panels in under four months.
Why It Matters for Trust & Control Assurance
- Continuous asset‑inventory and configuration‑management controls are designed to detect and remediate exactly this kind of rapid exposure before attackers can exploit vulnerable versions.
- Evidence of regular exposure monitoring feeds a defensible audit trail that maps to the “Asset Management” and “Vulnerability Management” objectives of NIST CSF 2.0.
- Verisq’s Control Mapping capability automates collection of exposure data, correlates it with internal inventories, and produces ready‑to‑use compliance evidence.
Who Is Affected – SaaS and workflow‑automation vendors (e.g., n8n), developers of self‑hosted AI agents, and operators of legacy industrial control systems that share internet‑facing services.
Recommended Actions
- Deploy continuous internet‑exposure scanning for all AI‑related assets.
- Prioritize patching or isolation of hosts running versions with critical CVEs.
- Integrate scan results into your configuration‑management database (CMDB) and map them to the relevant control objectives.
Technical Notes – The exposed n8n hosts largely run versions with CVE‑rated vulnerabilities ≥9.0; OpenClaw panels were publicly reachable within days of release. Older products (Jenkins, SharePoint, etc.) show minimal churn, indicating abandoned, unpatched instances remain online. Source: Help Net Security