Researchers from China’s national police academy have developed an artificial intelligence (AI) system capable of detecting illicit cryptocurrency transactions with nearly 90 percent accuracy, according to a study published in the Journal of Intelligence.
The research, conducted by the People’s Public Security University of China—an institution affiliated with the Ministry of Public Security—addresses growing concerns over the use of cryptocurrencies in illegal activities such as money laundering. Cryptocurrency transactions, while publicly recorded on blockchains, are pseudonymous and often cross international borders, complicating efforts by regulators to trace illicit flows.
The system employs a combination of advanced technologies, including a dynamic graph neural network, a memory mechanism, and a large language model (LLM), to analyse transaction patterns. The graph neural network maps the cryptocurrency ecosystem as an evolving network, studying the structure of transactions and the movement of funds over time. A memory module enables the AI to compare current transaction patterns with historical illicit activities, helping it detect emerging and adapting modes of criminal behavior. The LLM then translates these complex data points into natural-language risk assessments, providing not only binary classifications of legality but also risk scores and detailed reasoning to support regulatory decision-making.
In practical tests using the publicly available Elliptic Bitcoin transaction dataset—which features over 200,000 transaction nodes and 230,000 edges—the AI model achieved an overall accuracy of 89.4 percent. Its precision for identifying illicit transactions reached 89.1 percent, meaning nearly nine out of ten flagged transactions were confirmed illegal. However, the system’s recall stood at 64.5 percent, indicating it detected about two-thirds of all illicit transactions within the dataset.
Dr. Sun Jingchao, the study’s corresponding author and a specialist in criminal investigation and cybersecurity, described the framework as “a precise, generalisable and interpretable solution” that offers an “innovative technological pathway” for authorities confronting financial crime in the cryptocurrency sphere. The researchers emphasized that the memory component enhances the model’s ability to recognise new laundering methods, while the LLM contributes to generating transparent and traceable explanations for flagged transactions, boosting the system’s practical utility.
The development comes amid intensified efforts by Chinese authorities to clamp down on virtual currency-related offenses. In 2025, prosecutors indicted more than 3,200 individuals for money laundering linked to cryptocurrencies and underground banking networks, according to the Supreme People’s Procuratorate. Additionally, courts have dealt with large-scale cases involving billions of yuan in funds allegedly tied to online gambling and other illegal activities, underscoring the ongoing challenges in policing illicit financial flows in the digital asset space.
This AI-driven approach represents a significant advancement over traditional rule-based monitoring systems, which often struggle to keep pace with the evolving tactics employed by criminals in cryptocurrency markets. The research team advocates for the adoption of such intelligent frameworks to bolster regulatory oversight and enhance the detection and prevention of financial crimes.