Internet-Exposed AI Infrastructure Climbs 63% in Nine Months

Internet-exposed artificial intelligence infrastructure rose 63% over nine months to more than 294,000 public IP addresses, increasing the number of AI and large-language-model tools reachable from the public internet.

The comparison covers 43 identified AI and LLM tools, up from approximately 183,000 exposed IP addresses in October 2025. It was published July 23, 2026.

Langflow exposure increased 169% during the period. The AI application-development platform had 18 vulnerabilities recorded from 2024 through 2026, including 14 with CVSS scores above 8.0 and four entries in the Cybersecurity and Infrastructure Security Agency (CISA) Known Exploited Vulnerabilities catalog.

LiteLLM exposure rose 97%. The unified proxy for large-language-model providers is affected by CVE-2026-42208, a pre-authentication SQL-injection vulnerability listed in CISA’s Known Exploited Vulnerabilities catalog.

The vulnerability affects LiteLLM versions 1.81.16 through 1.83.6 and was fixed in version 1.83.7. It carries a CVSS v4 score of 9.3. An unauthenticated attacker could potentially read or modify proxy database data through a specially crafted Authorization header.

The figures measure publicly visible exposure through an internet-mapping dataset and related security research. They do not establish that the identified systems were compromised, nor do they represent the total number of AI systems in operation.

The increase highlights the expanding public reach of AI infrastructure while known security flaws remain present in some rapidly growing tools.