Researchers at China's national police university say they have built an AI framework that detects illicit Bitcoin transactions more accurately than mainstream large language models, including OpenAI's GPT-4o and Google's Gemini 3 Pro, according to the South China Morning Post, which reported the study achieved close to 90% overall accuracy against the Elliptic Bitcoin dataset, a public benchmark of about 200,000 transactions.
Chinese Police AI Beats US Models at Detecting Bitcoin Laundering
3 August 2026 - 09:53 UTC
By Oihyun Kim & Christina Pantin
The framework, called MDGNN-LLM, works in three stages: it maps the web of Bitcoin transactions as it evolves, checks new activity against a library of past laundering patterns, and asks an LLM to explain in plain language whether the case looks illicit.
Built on Alibaba's Qwen3 Model
The LLM at the core of the framework is Qwen3-14B, an open-source model released by Alibaba, which aligns with Beijing's stated push for technological self-reliance. The choice signals that Chinese law enforcement research on onchain forensics can now build its LLM layer entirely on domestic infrastructure.
Onchain analytics tools from Western vendors, such as Chainalysis and TRM Labs, still dominate that layer globally, with domestic Chinese alternatives such as OKG-affiliated OKLink's OnChain AML operating at a smaller scale.
Undergraduates, laptop deployment
The paper was authored by Zhu Rongxuan and Zhou Changhao, undergraduates born in 2005 and 2003 respectively, and corresponding author Sun Jingchao, a 32-year-old lecturer and doctoral candidate at the People's Public Security University of China (PPSUC), which is affiliated with the Ministry of Public Security. The undergraduate-led authorship indicates crypto forensic training is now reaching the bachelor's level at PPSUC, and isn't confined to specialist state labs.
All training and inference ran on a consumer gaming laptop with an AMD Ryzen 7 7840HS processor and an Nvidia RTX 4060-class GPU, hardware more typical of a student's PC than a research cluster.
A $3tn Crypto Market Contradiction
The paper opens by stating that as of January 2026, the authors put global cryptocurrency market capitalization above $3tn, calling crypto both a driver of financial innovation and a concealed channel for money laundering and fraud. The figure matches independent market data through January, an unusually direct acknowledgement of the sector's scale from a paper affiliated with the ministry that enforces Beijing's crypto trading ban, which it has repeatedly tied to money-laundering risk.
The study lands as Beijing continues to enforce its 2021 ban on crypto trading while building out the legal machinery to prosecute digital-asset cases. On 9 Mar 2026, Procurator-General Ying Yong told the National People's Congress that prosecutors indicted 3,259 people in 2025 for money laundering involving virtual currencies, underground banks, and other channels.
