Skip to content

README.md

Latest commit

 

History

History
115 lines (76 loc) · 2.46 KB

README.md

File metadata and controls

115 lines (76 loc) · 2.46 KB
Raw

AI Threat Modeling Assistant

A deterministic AI/ML threat modeling engine that identifies security risks across LLMs, traditional machine learning systems, and agentic AI workflows.

🌐 Live Tool: https://aimlthreats.com

Unlike LLM-based tools, this engine uses a rule-based approach to produce consistent, explainable, and audit-friendly threat models.

Why this tool?

AI systems introduce new and evolving attack surfaces:

  • Prompt injection
  • Data poisoning
  • Model extraction
  • Adversarial inputs
  • Agentic misuse and tool abuse

Most teams struggle to:

  • Identify relevant threats for their AI architecture
  • Map risks to compliance frameworks
  • Understand how to test and validate security controls

This tool bridges that gap using deterministic logic instead of black-box AI.

Key Design Principle

This tool is rule-based, not generative AI.

Why?

  • No hallucinated threats
  • Fully explainable logic
  • Consistent outputs
  • Suitable for audits and compliance
  • Deterministic behavior across runs

⚙️ Features

  • ✅ Supports multiple AI types:

    • LLMs
    • Traditional ML
    • Agentic AI systems

  • 🔍 Threat identification based on system inputs

  • 🧩 Mapping to security frameworks:

    • OWASP Top 10 (LLM & ML)
    • MITRE ATLAS
    • NIST AI RMF (CIA + Abuse)

  • 🔗 Attack chain detection (deterministic patterns)

  • 🧪 “How to Test” guidance for each finding

  • 📊 Severity scoring with contextual adjustments

  • 📄 Clean, structured report output

How it works

  1. User provides system details:

    • AI type (LLM / ML / Agentic)
    • Data sensitivity
    • Exposure (public/internal)
    • Controls (validation, logging, etc.)

  2. Rules engine evaluates:

    • Threat conditions
    • Control gaps
    • Risk scoring

  3. Findings are:

    • Mapped to OWASP / MITRE / NIST
    • Grouped into deterministic attack chains (when applicable)

  4. Output includes:

    • Severity (Low / Medium / High / Critical)
    • Description of risk
    • How to test
    • Mitigation guidance

Example Output

  • Scenario:

  • AI Type: LLM
  • Exposure: Public
  • Input Validation: No
  • Output Filtering: No
  • Sensitive Data: Yes

Findings:

  • Missing Input Validation → High
  • Missing Output Filtering → High
  • Sensitive Data Exposure Risk → High

Attack Chain (Triggered):

Chain: Untrusted Input → Unsafe Processing → Data Exposure