AI-driven Threat Modeling-as-a-Code (TaaC-AI)

It leverages AI to analyze service descriptions, identify security threats, visualize data flow and suggest remediations based on the STRIDE framework and OWASP guidelines. The script generates in-depth HTML report and includes feature for manual risk assessment.

Design

How It Works

• YAML File Processing: Loads and validates a YAML file containing the service details
• AI Threat Analysis: If an OpenAI API key is provided, the script uses AI to generate a comprehensive threat modeling analysis
• AI Threat Cross validation: Various LLMs performs a validation results
• Data Flow Generation: Automatically generates a visual representation of the data flow within the service
• Manual Risk Management: Users can manually add, modify, or cross out risks in the generated report
• Report Generation: Produces a detailed HTML report, including both AI-generated and manually added risks

Model supported

• GPT-3.5
• GPT-4
• Claude 3 Haiku
• Mistral 7b (through ollama)
• ‎Gemini (Planned)

Set API Key

Set the openai api key as an environment variable OPENAI_KEY and ANTHROPIC_KEY in your operating system

OpenAI API Anthropic API

Linux/Mac

export OPENAI_KEY=sk-ApiKeyExample

export ANTHROPIC_KEY=sk-ant-api03-ApiKeyExample

Windows

set OPENAI_KEY=sk-ApiKeyExample

set ANTHROPIC_KEY=sk-ant-api03-ApiKeyExample

Install dependencies

pip3 install -r requirements.txt

Install Mistral and Ollama

1. Download and install Ollama ollama.ai
2. Install Mistral

ollama pull mistral

3. Start ollama service (make sure the Ollama desktop app is closed)

ollama serve

Now you can use mistral as the main model:

python3 TaaC-AI.py --model mistral <path_to_yaml_file>

Or for cross-validation

python3 TaaC-AI.py --model claude --cross-validation mistral <path_to_yaml_file>

How to Use ❓

1. Create a valid service description using these guidelines
2. Execute the script (GPT-3.5 is used by default)

python3 TaaC-AI.py <path_to_yaml_file>

Use gpt-4, claude or mistral as a model to identify threats by specifying the --model option

python3 TaaC-AI.py --model gpt-4 <path_to_yaml_file>

To perform Threats result validation by another LLM use --cross-validation option.

python3 TaaC-AI.py --model claude --cross-validation claude <path_to_yaml_file>

3. Open generate .html report
4. Review/Edit AI-driven Threat Modeling Analysis table, and for false positives or resolved issues, mark the 'Status' checkbox
5. Add manually identified threats to the table (optional)
6. Download the report via the Download Report button

Usage Example 🏁

1. Valid service description example

Version: '1.0'
Date: 14.11.2023

# Authentication Service Description
Description:
  Name:  AuthService
  Type: Service
  Criticality: Tier1

# Service Functionality
Functionality: Handles user authentication, including login and token generation.

# Data Processing Details
DataProcessed: 
  Type: Confidential
  DataCategory: Auth
  EncryptionAtRest: Yes

# Components Used by the Service
Components:
  Internal: 
    Exist: Yes
    Source: Private
    Note: Scoped Package Access
  External: 
    Exist: Yes
    PackageManager: NPM

# Pipeline Configuration
Pipeline:
  Type: GithubActions
  CODEOWNERS: Yes
  BranchProtection: Yes
  SignCommits: Yes
  PinActions: Yes
  
# Network Information
Network:
  Access: Private

# Authentication Service Data Flow
dataFlow:  # Removed the dash here
  - name: UserAuthenticationFlow
    description: Authenticates users and issues tokens.
    source: UserLoginInterface
    EncryptionTransit: Yes
    Authentication:
      Exist: Yes
      Type: JWT
    Authorization: read-write
    Protocol: HTTPS
    Communication:
      Type: RESTful API
    interactions:
      - from: UserLoginInterface
        to: AuthService
        method: RESTful API
        protocol: HTTPS
      - from: AuthService
        to: UserDatabase
        method: Query
        protocol: JDBC
    servicesInvolved: [UserLoginInterface, AuthService, UserDatabase]

2. Script execution
3. Download and Review the generated HTML report

5. Add Threats manually

6. Download the latest report

Roadmap 🗓️

• Template Design
• Basic Functionality
• GPT-3 Integration
• Report generation
• Manually adding identified threats
• GPT-4 Integration
• Claude Integration
• LLM Cross Validation
• Mistral Integration via Ollama
• ‎Gemini Integration
• Accuracy Comparison

Contact 📧

All suggestions write to yevhsec1@gmail.com