AWS AI Security Maturity Model
A prescriptive security maturity model for AI and generative AI workloads on AWS. For each capability and maturity phase, it gives you tactical controls: which AWS service to use, how to configure it, and how to verify it.
How to use this model
This is a prescriptive security maturity model for AI and generative AI workloads on AWS. It gives you tactical, verifiable AWS security controls for each capability and maturity phase, and traces them to the OWASP LLM Top 10 for technical risk, the NIST AI RMF for governance, and the AWS Well-Architected Generative AI Lens security best practices. It distills the AWS AI Security Framework, the AWS SRA for AI, and the Generative AI and Agentic AI Security Scoping Matrices into one navigable, measurable guide. It measures control adoption and maturity, not compliance or certification.
The matrix: capabilities x maturity phases
The model maps the 6 AI capabilities from the AWS SRA for AI (model inference, customization, RAG, tools, agents, and AI applications) against 3 maturity phases (Foundational, Enhanced, Advanced). Each cell proposes tactical AWS security controls.
Controls are cumulative: each capability builds on model inference, and each phase builds on the previous one. You never start over.
The three questions this model helps you answer
- How are we advancing AI to production securely, and what's the cost of getting it wrong?
- What data can our AI access, and how is that being governed?
- How do we know our controls are working, and can we manage incidents?
Tactical controls, not a summary
This is the differentiator. Most AI security guidance stops at principles; here every control is specific and verifiable: which AWS service to use, how to configure it, and how to validate it, mapped to the OWASP LLM Top 10 risks it mitigates and the applicable GenAI scopes. The controls for tools and agents also address the agentic security dimensions (identity, memory, orchestration, boundaries, guardrails, and logging), which most models overlook.
Measure and prove your coverage
Use the self-assessment to track your adoption per control, capability, and phase, keeping everything in your browser and exporting it to a file when you need to. The coverage and traceability page shows how the controls map to the OWASP LLM Top 10 (technical risks), the NIST AI RMF functions (risk governance), the AWS Well-Architected Generative AI Lens security best practices, and the agentic security dimensions, so you can prove coverage across the technical, architectural, and governance axes.
It also maps the controls to the AWS Security Hub AI Security Best Practices standard: the automated CSPM checks that validate the matching controls in your account at runtime. This is a complementary validation layer, not a one-to-one equivalence: it confirms configuration posture (encryption, network isolation, gateway authorization) but does not cover the controls that are not machine-detectable, such as governance, human oversight, or prompt hardening.
Languages and sources
Available in English, Spanish, and Portuguese. Built entirely on official, public AWS and industry sources.