Skip to content

The "AttackModeler" framework is designed to model and analyze attacker behavior using OpenAI's GPT. It focuses on extracting information from previously modeled CTF reports, to create a comprehensive representation of attacks.

Notifications You must be signed in to change notification settings

marcobarca/AttackModeler

Repository files navigation

Alt Text

The "AttackModeler" framework is designed to model and analyze attacker behavior using OpenAI's GPT. It focuses on extracting information from previously modeled CTF reports, to create a comprehensive representation of attacks.

What is a StepModel

The StepModel is a json file that represents the attack as a list of actions (potentially divided into subactions).

Alt Text

The model is flexible, allowing the specialization of actions through the utilization of a Taxonomy to label each action.

Alt Text

An action can be more thoroughly described by incorporating Digital Artefacts, namely any type of digital trace that allows the reconstruction of events.

Alt Text

What is an Hidden Path

It is possible to view the list of actions as a sequence where each node is connected to the next through a simple temporal dependency. Thanks to the use of GPT-4, it is possible to automatically extract correlations between nodes that are not directly linked. The correlations are closely tied to the label and digital artifacts of each node.

An edge connecting two non-consecutive nodes is referred to as a Hidden Path.

Alt Text

Usage

  1. Add your GPT API KEY in AttackModeler/src/lib/gpt_api_call.py
  2. Run with: python3 example.py

System Overview

1. Attack File Construction

Alt Text

2. Attrubute Population

Alt Text

3. Attack File Analysis

Alt Text

4. Graph Example

Alt Text

5. Report Example

Alt Text

6. Hidden Paths File Example

Alt Text

7. Metodology Overview

Alt Text

About

The "AttackModeler" framework is designed to model and analyze attacker behavior using OpenAI's GPT. It focuses on extracting information from previously modeled CTF reports, to create a comprehensive representation of attacks.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages