AI Security Assessment Methodology
The same lifecycle used for every lab in the catalog and every AI security engagement — a structured path from discovery through ongoing monitoring, not a one-off penetration test.
The Lifecycle
Nine stages, applied to a specific AI system
Discover
Inventory the AI systems in scope — applications, models, RAG components, agents, tools and the data they touch.
Map
Map trust boundaries, data flows, identities and dependencies across the model, retrieval, agent and tool layers.
Threat Model
Identify realistic threat scenarios for the specific architecture — not a generic checklist applied uniformly.
Test
Structured testing against the threat model — prompt-level, retrieval-level, agent/tool-level and infrastructure-level.
Reproduce
Confirm findings are reliably reproducible before they are reported, with a documented, repeatable test harness.
Measure Impact
Assess realistic business and security impact — data exposure, privilege misuse, integrity loss — not theoretical severity alone.
Mitigate
Implement prioritized, architecture-level mitigations and secure-design changes, not just a list of findings.
Retest
Verify that mitigations actually close the gap once implemented.
Monitor
Carry the findings into ongoing detection — logging, alerting and SOC/SIEM integration for the behaviors that mattered.
What Gets Assessed
Every layer of an AI system, not just the model
Generative AI creates an interconnected attack surface — see The AI Attack Surface for the full picture. This methodology is applied consistently across every layer:
A decision is waiting for sign-off. Get an independent view first.
Tell me what is being decided and what is at stake. I will first say honestly whether my experience fits — then we talk about scope, not before.
Or connect on LinkedIn first.