Threat Modeling · Agentic AI Security
MAESTRO
Multi-Agent Environment, Security, Threat, Risk & Outcome
The 7-Layer Threat Modeling Framework for Agentic AI
Presented by Ken Huang
Fellow & Co-Chair of CSA AI Safety Working Groups · CEO, DistributedApps.ai
About Speaker
Ken Huang, CISSP
- 📚 AI Book Author & Speaker (Harness Engineering, MAESTRO)
- 🎓 Adjunct Professor, University of San Francisco
- 🛡️ OWASP AIVSS Project Lead
- 🤝 Fellow & Co-Chair of two CSA AI Safety Working Groups
- 🔟 Core Member, OWASP Top 10 for LLM Applications
- 🏛️ AIUC-1 Consortium Member
- 🔬 Grant Review Member for Schmidt Sciences
- 🧑🏫 EC-Council Instructor — Generative AI for Cyber Security
- 🚀 CEO of DistributedApps.ai
📚 AI & Academic Publications
Amazon Page ➔
The Problem
Why agentic AI needs a new threat model
Legacy frameworks & where they stop
| Framework | Focus | Methodology | Use cases |
|---|---|---|---|
| STRIDE | Six threat categories | Predefined categorization | Small teams, app/network |
| DREAD | Quantitative scoring | Numerical prioritization | Mature orgs |
| PASTA | Threats vs. business | Risk-centric, SDLC | Enterprises, compliance |
| OCTAVE | Operational risk | Asset-based, collaborative | Structured assessment |
| LINDDUN | Privacy threats | Privacy risk analysis | Healthcare, finance |
Where legacy assumptions fail for agents
The Framework
MAESTRO — a 7-layer threat modeling approach
Layer 1
Foundation Models
Layer 2
Data Operations
Layer 3
Agent Frameworks
Layer 4
Deployment Infrastructure
Layer 5
Evaluation & Observability
Layer 6
Security & Compliance
Layer 7
Agent Ecosystem
How to Apply It · Step 1
Layer mapping — mapping a system onto Anthropic MCP
How to Apply It · Step 2
Focus on the agentic factors
Within each layer, reason about the five properties that make agents fundamentally different from traditional software.
The Tooling
MAESTRO Sentinel maestro-sentinel.com ↗
Describe or upload your architecture — it maps threats across all seven layers, cross-referenced to OWASP AIVSS, MITRE ATLAS & the OWASP Agentic Top 10.
1 · Describe your architecture
2 · Threat intelligence library
3 · Per-threat analysis & mitigation
Connect
Let's talk agentic security
Ken Huang, CISSP — AI Book Author · Speaker · DistributedApps.AI · OWASP Top 10 for LLM Apps.
Scan the code to continue the conversation on threat modeling for agentic AI.
More info
MAESTRO Sentinel — the tool · maestro-sentinel.com ↗
DistributedApps.ai · distributedapps.ai ↗
MAESTRO framework — my CSA blog ↗
Run a MAESTRO analysis ↗