Dynamic Graph-based Fingerprinting of In-browser Cryptomining

TL;DR

This paper introduces a novel graph-based method for detecting in-browser cryptomining malware that is robust against obfuscation, using data-flow graphs and similarity measures to improve detection accuracy.

Contribution

The paper presents a new approach using instruction-level data-flow graphs and a graph comparison technique to detect cryptomining, outperforming existing methods especially under obfuscation.

Findings

High detection accuracy against obfuscation techniques

Effective graph simplification preserves key structures

Method applicable to various platforms beyond browsers

Abstract

The decentralized and unregulated nature of cryptocurrencies, combined with their monetary value, has made them a vehicle for various illicit activities. One such activity is cryptojacking, an attack that uses stolen computing resources to mine cryptocurrencies without consent for profit. In-browser cryptojacking malware exploits high-performance web technologies like WebAssembly to mine cryptocurrencies directly within the browser without file downloads. Although existing methods for cryptomining detection report high accuracy and low overhead, they are often susceptible to various forms of obfuscation, and due to the limited variety of cryptomining scripts in the wild, standard code obfuscation methods present a natural and appealing solution to avoid detection. To address these limitations, we propose using instruction-level data-flow graphs to detect cryptomining behavior. Data-flow graphs offer detailed structural insights into a program's computations, making them suitable for characterizing proof-of-work algorithms, but they can be difficult to analyze due to their large size and susceptibility to noise and fragmentation under obfuscation. We present two techniques to simplify and compare data-flow graphs: (1) a graph simplification algorithm to reduce the computational burden of processing large and granular data-flow graphs while preserving local substructures; and (2) a subgraph similarity measure, the n-fragment inclusion score, based on fragment inclusion that is robust against noise and obfuscation. Using data-flow graphs as computation fingerprints, our detection framework PoT (Proof-of-Theft) was able to achieve high detection accuracy against standard obfuscations, outperforming existing detection methods. Moreover, PoT uses generic data-flow properties that can be applied to other platforms more susceptible to cryptojacking such as servers and data centers.