AI-Powered Threat Modeling

Threat Technique Catalog for AWS: AWS Samples - Threat Technique Catalog

Threat modeling should be integrated directly into the design phase of the Secure Software Development Lifecycle (SSDLC).

Where Threat Modeling Fits in the SSDLC

$\text{Requirements}⟶\text{Design (Threat Modeling Happens Here)}⟶\text{Development}⟶\text{Testing}⟶\text{Deploy}⟶\text{Monitor}$

---

SSDLC Threat Modeling Timeline

1. Before Design

• Identify trust boundaries.
• Classify data sensitivity.
• Spec out security requirements.

2. During Design (Core Threat Modeling)

• Model System Architecture: Map components and interactions.
• AI-Assisted Threat Analysis: Let LLMs help discover edge-case threats.
• Map STRIDE per Component: Group threats into the STRIDE categories:
  • Spoofing identity
  • Tampering with data
  • Repudiation
  • Information disclosure
  • Denial of service
  • Elevation of privilege
• Define Mitigations Early: Document counter-measures before code construction begins.

3. After Design

• Threats Become Test Cases: Write automated security tests based on threats.
• Integrate into PR Reviews: Use threat models to guide pull request evaluations.
• Feed into Penetration Testing: Provide the threat catalog to penetration testers to focus their scope.

---

Tooling: Threat Composer AI & Kiro

• Threat Composer AI:
  • Powered by Bedrock Claude 3.5 Sonnet (or Sonnet 4) with pay-as-you-go token pricing (Beware of the cost!).
  • Uses 8 specialized agents working in parallel.
  • Equipped with a CLI and an MCP (Model Context Protocol) server.
  • Automatically parses CloudFormation templates or Terraform code to produce architecture and dataflow diagrams.
• Kiro CLI:
  • If you have an existing architecture diagram, you can upload it to Kiro to perform visual threat modeling and analyze data flows.

TIP

Start Small: A quick, 15-minute threat modeling session during design is infinitely better than no threat model at all. Don’t let perfection be the enemy of security.