Via theverge.com
The framework combines graph neural networks with large language models to flag suspicious Bitcoin transactions, and it already outperforms existing detection tools.
Researchers at China’s top police university have built an AI system that can identify illicit Bitcoin transactions with 89.4% overall accuracy, a development that could reshape how law enforcement agencies worldwide approach crypto-related financial crime.
The team at the People’s Public Security University of China published their findings in May in the Journal of Intelligence. The South China Morning Post reported on the breakthrough, highlighting how the framework outperforms existing mainstream baseline models for detecting dirty money flowing through blockchain networks.
How the AI actually works #
This new framework combines three technologies: dynamic graph neural networks that map the 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 tested their system on the Elliptic Bitcoin dataset, a public benchmark that contains 203,769 transaction nodes and 234,355 edges connecting them. The model achieved 89.1% precision and a 64.5% recall rate specifically for illicit transactions.
When the model flags something as suspicious, it’s right about 89% of the time. But it only catches about 65% of all illicit transactions in the dataset, meaning roughly one-third of bad transactions still slip through undetected.
Why explainability matters more than raw numbers #
Dr. Sun Jingchao, the study’s corresponding author, pointed to the system’s historical pattern matching capabilities as a key advantage. The memory module lets the AI reference past illicit transaction patterns and apply reasoning chains, generating risk scores alongside its classifications.
The framework also produces natural language explanations for its decisions. Instead of just spitting out a probability score, it can articulate the logic behind its assessment. For regulators and prosecutors who need evidence that holds up in court, that’s a meaningful upgrade over a confidence interval.
China’s broader crackdown on crypto crime #
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, roughly $444 million, linked to online gambling debts.
The AI framework’s concentrated focus on Bitcoin is notable. The research doesn’t mention specific protocols, companies, or other tokens, reinforcing Bitcoin’s historical role as the default rail for illicit transactions.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our