Application of Game Theory For Active Cyber Defense Against Advanced Persistent Threats

Abstract

Advanced persistent threats (APTs) are determined, adaptive, and stealthy threat actors in cyber space. They are often hosted in, or sponsored by, adversary nation-states. As such, they are challenging opponents for both the U.S. military and the cyber-defense industry. Current defenses against APTs are largely reactive. This thesis used machine learning and game theory to test simulations of proactive defenses against APTs. We first applied machine learning to two benchmark APT datasets to classify APT network traffic by attack phase. This data was then used in a game model with reinforcement learning to learn the best tactics for both the APT attacker and the defender. The game model included security and resource levels, necessary conditions on actions, results of actions, success probabilities, and realistic costs and benefits for actions. The game model was run thousands of times with semi-random choices with reinforcement learning through a program created by NPS Professor Neil Rowe. Results showed that our methods could model active cyber defense strategies for defenders against both historical and hypothetical APT campaigns. Our game model is an extensible planning tool to recommend actions for defenders for active cyber defense planning against APTs.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 01, 2023
Accession Number
AD1224525

Entities

People

  • Andrew S. Benn
  • Stephanie M. Benn

Organizations

  • Naval Postgraduate School

Tags

Fields of Study

  • Computer science

Readers

  • Artificial Intelligence
  • Computational Modeling and Simulation
  • Cybersecurity.

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy
  • AI & ML - Neural Networks
  • Cyber
  • Space