Exploring Applications of NLP to Create Deceptive Sandbox Environments with Honeyfiles

Abstract

Various techniques are in practice to create a sandboxed environment to study malicious actors and/or divert them from gaining access to critical assets. Honeypots and honeynets have long been used to create decoy systems that lure adversaries to attack them, thereby revealing tactics, techniques, and procedures. When mimicking enterprise systems and networks, certain artifacts will have to be strategically placed so that they look organically generated, yet do not contain confidential data. Text assets such as word documents, audit reports, and email conversations are usually targeted for their potential value. This research explores the possibilities of generating fake, yet convincingly real documents based on relevant publicly available data. The research gaps in the existing state-of-the-art and appropriate technologies to address them would be evaluated. The project would also include proposed empirical research to verify if these NLP-generated documents are better than random gibberish text.