Immaturity & Moral Hazard in the Cyber Insurance Market
Topic: Artificial Intelligence Research for Forecasting Exploit Usage
Vulnerability disclosure rates are at an all-time high – averaging over 1,000 per month in 2019 – more than twice as much as in 2016. But while disclosure rates have remained at this high level, hackers still only exploit a small fraction ranging from 2%-3% by most studies. Ironically, the fact that exploited vulnerabilities make up such a small portion mean that this a particularly challenging machine learning problem. In this talk will review a series of peer-reviewed research papers that were produced under U.S. government grant funding that have investigated this problem. Through a combination of machine learning, graph theory, and data mining (from sources including social media, deepweb, open web, and Tor sites), these approaches provided promising results. These techniques leveraged an understanding of not only the content of hacker discussions, but also the underlying social structure of these communities as well as technical information about the vulnerabilities themselves. This, in-turn, enabled successful forecasting of exploits before they become available – providing a 20-fold improvement in terms of precision. This talk not only reviews the peer reviewed research, but also gives insight into how machine learning can be used to address cybersecurity problems and provides examples of exploit usage successfully predicted ahead of time.