Researchers at the People's Public Security University of China, the country's leading police academy, have developed an artificial intelligence system capable of identifying illicit Bitcoin transactions with 89.4% overall accuracy. The advancement could influence how law enforcement agencies worldwide investigate cryptocurrency-related financial crime, an area where blockchain analytics firms such as Chainalysis, Elliptic, and TRM Labs already provide transaction-monitoring tools to government agencies across multiple jurisdictions.

The team published their findings in May in the Journal of Intelligence. The South China Morning Post reported on the research, noting that the framework outperforms existing mainstream baseline models for detecting illicit funds moving through blockchain networks.

Architecture and Methodology

The framework integrates three technologies: dynamic graph neural networks that map evolving relationships between transaction nodes, a memory module that stores historical patterns of illicit activity, and large language models that handle the reasoning and classification layer.

The researchers evaluated the system using the Elliptic Bitcoin dataset, a public academic benchmark created by UK-based blockchain analytics firm Elliptic, comprising 203,769 transaction nodes and 234,355 edges connecting them. The dataset labels transactions as illicit, licit, or unknown based on real-world case data and has been widely used in cryptocurrency crime research since its release in 2019.

The model achieved 89.1% precision and a 64.5% recall rate specifically for illicit transactions. In practical terms, when the model flags a transaction as suspicious, it is correct approximately 89% of the time. However, it captures only about 65% of all illicit transactions in the dataset, leaving roughly one-third undetected. That recall gap means a meaningful portion of criminal fund flows could still pass through undetected, which matters for investigators who cannot rely on a single tool to surface all suspicious activity.

Explainability and Prosecutorial Value

Dr. Sun Jingchao, the study's corresponding author, identified the system's historical pattern matching capabilities as a key advantage. The memory module enables the AI to reference past illicit transaction patterns, apply reasoning chains, and generate risk scores alongside its classifications.

The framework also produces natural language explanations for its decisions. Rather than outputting only a probability score, the system can articulate the reasoning behind each assessment. For regulators and prosecutors who require evidence admissible in court, this represents a significant improvement over a standard confidence interval. The explainability feature aligns with a broader trend in AI governance frameworks, including the European Union AI Act adopted in 2024, which emphasizes transparency and auditability requirements for systems used in law enforcement contexts.

China's Enforcement Context

The research emerges amid an intensifying crackdown on cryptocurrency-related crime in China, which banned domestic cryptocurrency trading and mining in 2021. Despite the ban, cross-border crypto-linked money laundering has persisted, driving demand for more sophisticated detection capabilities.

In March 2025, Chinese prosecutors indicted 3,259 individuals involved in money laundering activities connected to virtual currencies and underground banking operations.

On July 25, 2026, a Chinese court handled a case involving nearly 3 billion yuan, approximately $444 million, linked to online gambling debts.

The AI framework's exclusive focus on Bitcoin is notable. The research does not reference specific protocols, companies, or other digital assets, underscoring Bitcoin's persistent role as the primary conduit for illicit transactions in cryptocurrency markets.