Deep Graph Networks for Analysing Illicit Activity in Cryptocurrency to Combat Financial Cybercrime

Financial cybercrime is an increasing threat to the public, corporations, and government entities as billions of euros are stolen and laundered annually through financial networks. Artificial intelligence methods offer investigators a method of defending against illicit activity, preventing loss of assets, and identifying cybercriminals. Cryptocurrency offers attackers unique avenues to launder their illicit funds through the decentralised networks. Cryptocurrency crime is estimated to be worth over \$24 billion in 2023 and is steadily rising year-on-year with Bitcoin being the most popular cryptocurrency of choice for criminals. Current techniques deployed to track and prevent Bitcoin related crime are being evaded by advanced and sophisticated tooling created by privacy advocates and cybercriminals. The emerging threats from advanced cybercriminals and malicious users of Bitcoin emphasise the requirement for a new all encompassing approach that tackles varying levels of obfuscation on the blockchain. This thesis addresses the issues of illicit detection evasion by introducing FraudLens, a framework for identifying illicit activity in Bitcoin. FraudLens is established to (i) address the evolving threat landscape within cryptocurrency and identify and explain the wide variety of methods to disguise and obfuscate one self on the blockchain network, (ii) introduce graph restructuring methods to improve graph neural network performance in transaction monitoring and the broader node imbalance classification task, (iii) integrate large language models as a method of explainable artificial intelligence through contextual narrative generation and similarity analysis, and (iv) perform topological data analysis on transaction graph structures to identify persistent features to enhance the robustness of transaction monitoring demonstrated by defending against adversarial graph neural network attacks. The results across the three experiments demonstrate a holistic approach to tackling financial cybercrime in cryptocurrency and significantly push the state-of-the-art by addressing the advanced nature of the current Bitcoin threat landscape.