Chinese police researchers have created an AI-driven system capable of identifying illicit Bitcoin transactions with nearly 90% accuracy. This breakthrough could significantly impact how authorities track and combat crypto crime across the globe.
The People’s Public Security University of China unveiled the technology following a study published in the Journal of Intelligence. Their AI model outperforms existing tools by combining dynamic graph neural networks, memory modules storing historic illicit patterns, and large language models that enhance reasoning and classification of suspicious activity.
How the AI Spots Illicit Transactions
The system was put to the test using the Elliptic Bitcoin dataset, which includes over 200,000 transaction nodes and more than 230,000 connections between them. It achieved an 89.1% precision rate, meaning it is right almost nine times out of ten when it flags a transaction as suspicious. However, it catches around 65% of all illicit transactions in the dataset. That leaves a significant portion roughly a third of criminal activity undetected.
One standout feature is the AI’s memory module, which allows it to reference past patterns of illicit transactions and reason through them, producing clear risk scores. Instead of just offering a confidence score, the system can produce natural language explanations, making its logic easier to interpret. This could prove vital for prosecutors and regulators who require transparent evidence in court.
This advancement arrives amid China’s intensified crackdown on crypto crime, which saw over 3,000 indictments in early 2025. Tools like this AI could reshape enforcement strategies and set new standards worldwide.
