Independent Agent Boundary Reviews.

What it is. An independent, adversarial look at what your AI agent can actually do versus what your documentation says it can do: the execution boundary, the authority a tool call inherits, what happens when the input comes from somewhere you didn't expect. Delivered as a written report with a claims table: what's genuinely demonstrated, what's only described, and what would still stop a buyer from trusting the claims.

Who this is for. Teams shipping an agent product, fielding hard security questions from an enterprise buyer, who need an answer that isn't "trust us." Not a general security audit: the boundary the agent operates inside, specifically.

I find the security holes in AI agent infrastructure, and I have filed them: four findings across two third-party MCP servers, three scored CVSS 8.1 to 9.1 by my own assessment, coordinated disclosure in progress; two filed with MITRE CNA-LR 2026-08-17