Proactive Vulnerability Assessment via AI

1
Proactive Vulnerability Assessment via AI
Alan Fern and Thinh Nguyen, EECS, Oregon State
Universityhttp//web.engr.oregonstate.edu/cybert
rust
Alan Fern

• Scope Utilize artificial intelligence (AI)
  techniques to discover new network
  vulnerabilities before attackers do. Current
  focus is on DDoS attacks of P2P networks.

Intelligent Agent
uses interaction with the network and AI
planning to generate novel attack strategies

• Potential Applications
• Discover unknown vulnerabilities of a particular
  network under particular security mechanisms
• Investigate worst-case performance of newly
  proposed defense mechanisms, ideally giving
  provable worst-case damage bounds
• Investigate the potential risk from malicious
  use of AI techniques

• Our General Approach
• Model a network with security mechanisms as a
  Markov decision process
• Use simulation-based planning to uncover an
  attacker policy to maximize network damage
• Provide provable guarantees on achievabledamage
  by extending theoretical results from spread of
  influence research

Peer to Peer Network
Preliminary Results DDoS P2P Attacks

• Developed fast simulator for large scale
  P2Pnetworks with several security mechanisms
• Developing a framework for optimization of DDoS
  attack policies relative to a given network
• Initial results show improved attacks compared to
  a random DDoS attack and an existing heuristic
• Developing theory based on sub-modular functions
  to guarantee approximately optimal attacks are
  returned