What is it about?
AI agents are different from traditional software because they can reason about goals, make decisions, use tools and APIs, access data, retain context, and take actions with limited human involvement. These capabilities create security risks that conventional threat-modeling approaches may not fully capture. This article explains a practical approach to threat modeling agentic AI systems. It shows security teams how to examine an agent’s goals, identity, permissions, tools, memory, data access, trust boundaries, and runtime behavior. It also considers how risks can emerge when an agent’s plans, permissions, context, or actions change during execution. The goal is to help architects, developers, security practitioners, and organizations identify agent-specific threats earlier and design security controls that remain effective while AI agents are operating.
Featured Image
Photo by Igor Omilaev on Unsplash
Why is it important?
AI agents are moving from systems that simply generate answers to systems that can take real actions. They may call APIs, execute tools, access enterprise data, interact with other agents, and make decisions across multi-step workflows. This changes the security problem. Security reviews cannot focus only on prompts, models, or application boundaries. They must also consider what an agent is authorized to do, how its authority changes during execution, what tools and resources it can reach, and whether its actual actions remain aligned with the original intent. This article provides practitioners with a structured way to reason about these risks through agentic threat modeling. The approach can help organizations identify dangerous trust boundaries, excessive authority, unsafe tool interactions, and runtime risks before they become security incidents. As autonomous AI becomes more deeply integrated into enterprise systems, threat modeling that accounts for agent behavior and runtime authority can become an important part of secure AI engineering.
Perspectives
I wrote this article because I believe AI agents require us to rethink some assumptions behind traditional application security. When software primarily responds to requests, we can often reason about security around relatively stable application flows and trust boundaries. An AI agent may instead interpret a goal, create a plan, select tools, access resources, and take a sequence of actions that was not explicitly defined in advance. For me, the important security question is therefore not only “Can this agent perform this action?” but also “Should this agent be allowed to perform this action, in this context, at this point in the workflow, for this purpose?” My goal with this work is to make agentic AI security practical for architects, developers, threat modelers, and security teams. I hope it encourages practitioners to examine identity, authority, tools, memory, trust boundaries, and runtime behavior together rather than treating AI security as only a model or prompt-security problem.
Ravindra Annam
Read the Original
This page is a summary of: Running an Agentic Threat Model, Queue, September 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3848521.
You can read the full text:
Resources
Running an Agentic Threat Model — ACM Queue
Official ACM Queue publication of “Running an Agentic Threat Model” by Ravindra Annam. The article presents a practical approach to identifying and analyzing security risks in agentic AI systems, including identity, authority, tools, memory, trust boundaries, and runtime behavior.
Breaking AI: Real-World Attacks & Defenses for Agentic AI Systems
Invited university presentation by Ravindra Annam on security threats and defenses for agentic AI systems. The session examines how autonomous AI agents introduce security challenges involving identity, authority, tool use, data access, trust boundaries, and runtime behavior, extending practical themes addressed in “Running an Agentic Threat Model.
Securing AI Agents: Guardrails and Runtime Controls for the Autonomous Enterprise
Recorded ISC2 Think Tank panel featuring Ravindra Annam and fellow cybersecurity practitioners discussing the security of enterprise AI agents. The session examines agent identity and permissions, tool and MCP interactions, data exposure, behavioral controls, policy enforcement, monitoring, and runtime guardrails—topics closely related to the agentic AI threat-modeling challenges explored in “Running an Agentic Threat Model.”
AI agents need their own identity before they need a gateway
Related article by Ravindra Annam examining identity, attribution, delegated authority, and runtime trust for autonomous AI agents. It complements “Running an Agentic Threat Model” by exploring why authentication alone is insufficient when agents can dynamically invoke tools, access enterprise data, use memory, and execute actions across multi-step workflows.
SecuriAI Agent Shield — AI Agent Security & Threat Modeling
AI security project created by Ravindra Annam exploring practical threat modeling and runtime security for agentic AI systems. The project includes AI-focused threat modeling workflows, agent risk detection, secure architecture analysis, runtime security concepts, and enterprise security visibility, providing a practical implementation context for themes discussed in “Running an Agentic Threat Model.”
Agentic AI Security Research & Threat Modeling — Ravindra Annam
Research hub for Ravindra Annam’s work on agentic AI security, including runtime authority-aware threat modeling, the AI Runtime Threat Matrix, runtime security models, agentic AI attack paths, and enterprise AI security controls.
The First 24 Hours of an AI Agent Security Incident
Related article by Ravindra Annam translating agentic AI security concepts into an operational incident-response approach. It examines identity-first containment, preservation of agent decision evidence, authority and access controls, and safe recovery when an autonomous agent is compromised, manipulated, or acts outside its intended boundaries. It complements “Running an Agentic Threat Model” by connecting proactive threat modeling with real-world agent incident response.
Practical Lessons from Deploying AI Securely at Scale
Related article by Ravindra Annam examining practical security lessons from enterprise AI deployments. It focuses on risks that emerge when AI systems receive permission to act across real business processes, emphasizing authority, access, runtime behavior, and security controls that keep autonomous systems aligned with business intent. These operational lessons directly complement the threat-modeling approach presented in “Running an Agentic Threat Model.”
Ravindra Annam — ResearchGate Research Profile
Research profile of Ravindra Annam featuring publications and research related to agentic AI security, AI-agent threat modeling, runtime security, cybersecurity, and secure autonomous systems.
Ravindra Annam — ORCID Researcher Profile
ORCID researcher profile for Ravindra Annam, providing a persistent researcher identifier and a record connecting scholarly publications and professional research activities.
Ravindra Annam — Google Scholar Profile
Google Scholar profile for Ravindra Annam providing a consolidated view of scholarly publications, citations, and research impact across agentic AI security, AI security, cybersecurity, and related areas.
Runtime Authority-Aware Threat Modeling for Agentic AI Systems
Related scholarly work by Ravindra Annam examining threat modeling for agentic AI systems through the lens of runtime authority, delegated permissions, tool use, trust boundaries, and autonomous execution. It provides additional research context for the practical agentic threat-modeling concepts discussed in “Running an Agentic Threat Model.”
Contributors
The following have contributed to this page







