Learn
Learning path
6 steps from the building blocks of an agent to a working threat model. Read a few pages, then pass a short quiz to complete each step.
- 6 steps
- 21 readings
Step 1: Component analysis
Learn the 6 key components of an agentic system (KC1–KC6) and where their security boundaries sit.
- Key components: All 6 key components and their sub-components.
- KC2 · Orchestration (Control Flow): Mechanisms that dictate the agent's overall behavior, information flow, and decision-making processes.
- KC4 · Memory Modules: Enable the agent to retain short-term (immediate context) and long-term information (past interactions, knowledge) for coherent and personalized interactions.
- Architecture patterns: 5 reference architectures built from those components.
Step 2: Threat mapping
Map threats to components using the catalog of 17 threats and 103 attack vectors, then see how they line up with the OWASP Agentic Top 10.
- Threat catalog: 17 threats, 103 attack vectors, rated by severity.
- T1 · Memory Poisoning: 6 attack vectors and 7 mapped controls.
- OWASP Top 10 for Agentic Applications: The 10 ASI entries and their real-world attack scenarios.
- ASI01 · Agent Goal Hijack: 4 attack scenarios and 8 prevention guidelines.
Step 3: Control selection
Pick controls from 20 mitigations organised by lifecycle phase (design, build, operations).
- Controls: 20 mitigations mapped to the threats they reduce.
- M7 · Zero Trust & Least Privilege Access: Mitigates 4 threats.
- Implementation by phase: The same controls grouped into design, build and operations work.
- Security checklist: A step-by-step checklist with progress tracking.
Step 4: AISVS verification
Verify the implementation against OWASP AISVS 1.01: 213 requirements in 12 chapters, at three assurance levels.
- OWASP AISVS 1.01: 213 requirements: 57 at L1, 110 at L2, 46 at L3.
- C9 · Orchestration & Agentic Security: 37 requirements in 6 sections.
- C10 · Model Context Protocol (MCP) Security: 39 requirements in 4 sections.
Step 5: NIST alignment
Connect the technical controls to governance: 55 of 72 NIST AI RMF subcategories map to an AISVS chapter on this site.
- NIST AI RMF mapping: 72 AI RMF subcategories mapped to AISVS chapters.
- Cross-framework taxonomy: OWASP Agentic Top 10, Cisco, AIVSS and AIUC-1 side by side.
- References: The standards and papers behind every mapping.
Step 6: Model it
Put it together: draw an agentic system in the Threat Modeler, place it on the MAESTRO layers and generate its threats and controls.
- Threat Modeler guide: Components, MAESTRO layers, analysis and exports.
- Sequential Agent Architecture: 5 key threats and 7 controls for this pattern.
- Open the Sequential Agent template: Start the Threat Modeler from a ready-made template.