DARD: Deceptive Approaches for Robust Defense Against IP Theft

With the rise of smart working and recent global events, the risk of cyberattacks is increasing steadily. Sometimes adversaries focus on stealing valuable data, such as intellectual property (IP): they exfiltrate a large volume of IP documents from a target company. They then identify those of their interest by leveraging automated methods. This work proposes the DARD (Deceptive Approaches for Robust Defense against IP theft) system, a framework designed to deceive adversaries who rely on automatic approaches to classify exfiltrated documents. Starting from an original repository of documents, DARD automatically generates a new deceptive repository that misleads popular automatic approaches, resulting in clusters of documents that are significantly different from the actual ones. By utilizing this approach, DARD aims to hinder the accurate clustering and the identification of the topic of documents by adversaries relying on automated techniques. The paper presents four deceptive operations (Basic Shuffle, Shuffle increment, Shuffle reduction, and Change topic) that DARD leverages to create a deceptive repository. We evaluate the efficacy of our approach by considering three different types of adversaries, each possessing varying levels of knowledge and expertise. Through extensive experiments, we show that the DARD system can deceive both automatic topic modeling and document clustering techniques, including widely-used commercial tools such as Amazon Comprehend. Hence, our solution provides a robust defense mechanism against Intellectual Property (IP) theft.