Capsule Research / 2026
The State of AI Agent Security
The broadest independent security audit of the agentic AI ecosystem to date. Built on internet-wide scanning, large-scale code analysis, and original zero-day research. A data-driven picture of how AI agents are being built, deployed, and left unprotected.
402,599exposed AI agent hosts
on the public internet
on the public internet
206,435agent skill files
analyzed
analyzed
164,692code files
statically audited
statically audited
