Two builders, four cold brews, and one urgent conversation about MCP security.
About the event
The security boundary for AI just moved. When agents were passive assistants, the question was what does the model say? Now agents take actions — irreversible, machine-speed, often against production systems — and the question is what does the model do, and who's watching when it does it?
Herman Errico (Senior PM at Vanta) wrote the spec for this. It's called AARM — Autonomous Action Runtime Management — and as of last month it lives at the Cloud Security Alliance, with Herman chairing the working group. Jiquan Ngiam built MintMCP as one of the four reference architectures AARM defines for putting it into practice.
So we're pouring four cold brews and letting them work through what it actually means to secure an agent at runtime.
Brew 1 — Why runtime is the new boundary. Logs, perimeter defense, post-hoc forensics — the entire security stack assumes you can clean up afterward. Agents break that assumption. Herman walks through why the action layer is where the controls have to live now. Cold brew we'll pair it with: Equator Coffees Organic Smooth Black.
Brew 2 — What AARM actually says. Intercept the action before it executes. Accumulate session context. Evaluate against policy and intent alignment. Enforce a decision — allow, deny, modify, defer. Record a tamper-evident receipt. Five moves, one loop, applied to every tool call an agent makes. Cold brew we'll pair it with: La Colombe Medium Roast.
Brew 3 — How Mint implements it. AARM defines four reference architectures (protocol gateway, SDK instrumentation, eBPF, vendor integration); MintMCP is the protocol gateway path. Jiquan shows how Virtual MCPs become the enforcement point — role-scoped, observable, replacing the current "every agent connected to everything with no audit trail" pattern. Cold brew we'll pair it with: STōK Espresso Blend Dark Roast.
Brew 4 — Monday morning. A four-step playbook — Discover, Assess, Prioritize, Enforce — for getting from wherever you are now to something you'd be comfortable showing an auditor.
You'll leave understanding AARM as a framework, not a vendor pitch, plus a concrete picture of what it looks like in production. Cold brew we'll pair it with: Chameleon Organic Original Concentrate.
About the Speakers
Jiquan Ngiam, CEO & Co-Founder, MintMCP: Jiquan is the CEO & Co-Founder of MintMCP, an AI agent governance platform that lets teams build securely on AI platforms. He is co-author of "Securing the Model Context Protocol: Risks, Controls, and Governance" and runs 20+ AI agents on MintMCP's product platform, giving him firsthand experience navigating the security challenges of production-grade agent deployments.
Herman Errico, Senior Product Manager, Vanta: Herman is a Senior Product Manager at Vanta focused on security frameworks, technical controls, and risk. He is co-author of "Securing the Model Context Protocol: Risks, Controls, and Governance" and the creator of AARM (Autonomous Action Runtime Management), an open specification defining how to secure AI agent actions at runtime. An IEEE Senior Member and patent holder, Herman brings deep expertise across ISO 42001, NIST AI RMF, and enterprise AI governance.
By submitting your information, you agree to share your data with and be contacted by MintMCP about their services. Your information will be processed in accordance with our privacy policy.