Generative Adversarial Networks from a Cyber Intelligence perspective

Abstract

Generative adversarial networks (GAN) are a hot topic in cyber intelligence, as they begin to demonstrate abilities that will assist the public intelligence analyst to play a more active role in global security. Not only can they help you sort through the vast OSINT sources of material to identify potential threats, but additional features are also now being demonstrated which will allow links to be drawn in real time between potential threats. This is an incremental report produced for the NATO Cooperative Cyber Defence Centre of Excellence, Cyber Intelligence division, under agreement. It focuses on preliminary work addressing a wider project to produce a paper on the cyber intelligence uses of GAN. The work has two main focuses: • to investigate different applications to assess the most interesting and promising in respect of cyber intelligence; and • to determine a framework to assess current examples of GAN that offer solutions relevant to cyber intelligence. To facilitate this, some questions need to be addressed: (1) what is a GAN? (2) what is cyber intelligence? (3) how do the two interact? The focus of the project is to assess the potential use of GAN in the context of cyber intelligence, the unified kill chain model and how ready such technology is for deployment; both for legal uses and illegal. This raises certain questions: (1) How might GAN be used to assist in the collection, validation, exploitation or reporting of intelligence information? (2) How might GAN be used to assist in gaining initial footholds in systems, propagate or pivot and finally achieve actions on an objective? (3) How has GAN has been used traditionally, what opportunities does this present to the intelligence community (IC) and are these uses legal? This focus is driven by two approaches, those that are collection-orientated and those that are offensive or defensive towards the cyber threat. The framework we will design will incorporate an assessment for both.