References & Citations
Computer Science > Cryptography and Security
Title: Software Updates Strategies: a Quantitative Evaluation against Advanced Persistent Threats
(Submitted on 16 May 2022 (this version), latest version 25 May 2022 (v2))
Abstract: Software updates reduce the opportunity for exploitation. However, since updates can also introduce breaking changes, enterprises face the problem of balancing the need to secure software with updates with the need to support operations. We propose a methodology to quantitatively investigate the effectiveness of software updates strategies against attacks of Advanced Persistent Threats (APTs). We consider strategies where the vendor updates are the only limiting factors to cases in which enterprises delay updates from 1 to 7 months based on SANS data. Our manually curated dataset of APT attacks covers 86 APTs and 350 campaigns from 2008 to 2020. It includes information about attack vectors, exploited vulnerabilities (e.g. 0-days vs public vulnerabilities), and affected software and versions. Contrary to common belief, most APT campaigns employed publicly known vulnerabilities. If an enterprise could theoretically update as soon as an update is released, it would face lower odds of being compromised than those waiting one (4.9x) or three (9.1x) months. However, if attacked, it could still be compromised from 14% to 33% of the times. As in practice enterprises must do regression testing before applying an update, our major finding is that one could perform 12% of all possible updates restricting oneself only to versions fixing publicly known vulnerabilities without significant changes to the odds of being compromised compared to a company that updates for all versions.
Submission history
From: Giorgio Di Tizio [view email][v1] Mon, 16 May 2022 15:33:36 GMT (498kb)
[v2] Wed, 25 May 2022 09:30:38 GMT (498kb)
Link back to: arXiv, form interface, contact.