Presentation Material

Abstract

Designing a game between an adversary and defender has been a challenge because the adversary’s tactics and strategies are unobservable. However, if the defender is powered by AI-based tools to get observations about adversary actions and tactics then a realistic game can be played. This talk shares a research on game theory and AI for information security.

AI Generated Summary^{may contain errors}

Here is a summary of the content:

The speaker, A and another person Suresh are discussing artificial intelligence (AI) in the context of security. The speaker explains that AI is an overarching umbrella of techniques that leverage neural networks to mimic human behavior. In security, AIs, , it’s difficult to define the state space due to its complexity.

To address this challenge, the speaker proposes breaking down the problem into subproblems using a stepwise process and defining the state space by referencing the Mitre kill chain attack matrix. They show examples of how this approach can be effective in detecting attacks.

Suresh asks about the percentage of human intelligence that AI will match up to, and the speaker clarifies that they are not trying to replicate human intelligence but rather define a step towards AI in security. The goal is to break down the problem into manageable parts, define the state space, and then work towards simulating real-world scenarios.

The speaker also mentions open-source tools like Caldera, which can simulate attackers based on the Mitre attack matrix. They emphasize that their approach is an incremental step towards achieving AI in security.

Finally, a moderator wraps up the session, inviting attendees to ask further questions offline and announcing a break for lunch.