Performance and Accuracy Metrics

Chinese researchers from the People’s Public Security University of China have developed an AI-driven framework that reportedly achieves nearly 90% accuracy in detecting illicit Bitcoin transactions. The tool, detailed in research published around August 2026, utilizes a hybrid architecture of Large Language Models (LLMs) and temporal memory modules to identify complex money-laundering patterns that traditional tracking methods often miss [Source: https://www.scmp.com/news/china/science/article/3272844/chinese-police-ai-algorithm-tracks-bitcoin-money-laundering-90-accuracy].

Performance and Accuracy Metrics

The tool is designed to combat "pseudonymous cross-border illegal trading," a major concern for Chinese authorities following the 2021 ban on cryptocurrency transactions.

Methodology and Technical Capabilities

The framework moves beyond simple heuristic analysis (tracking address-to-address hops) by incorporating advanced machine learning components:

• Memory Module: This component retains and compares historical transaction patterns, allowing the AI to recognize evolving laundering tactics over time [Source: https://www.scmp.com/news/china/science/article/3272844/chinese-police-ai-algorithm-tracks-bitcoin-money-laundering-90-accuracy].
• Large Language Model (LLM) Integration: The LLM processes transaction metadata and associated textual data to provide "interpretable solutions," helping human investigators understand why a specific transaction was flagged [Source: https://www.scmp.com/news/china/science/article/3272844/chinese-police-ai-algorithm-tracks-bitcoin-money-laundering-90-accuracy].
• De-anonymization via KYC: Authorities augment the AI's findings by retrieving Know Your Customer (KYC) data from major exchanges like Binance, OKX, and HTX through formal legal assistance channels to link flagged wallets to real-world identities [Source: https://cryptorank.io/news/feed/789456-chinese-police-detail-crypto-tracking-methods].

Known Limitations and Challenges

While the 90% accuracy rate is high, the tool faces several technical and legal hurdles:

• False Positives: Deep learning models in cryptocurrency tracking are known to have higher rates of false positives, which can complicate legal proceedings [Source: https://www.mdpi.com/journal/ai/special_issues/AI_Cryptocurrency].
• Privacy-Enhancing Technologies (PETs): The use of privacy coins (e.g., Monero), shielded transactions (Zcash), or cross-chain bridges remains a significant obstacle to maintaining tracking continuity [Source: https://www.scmp.com/news/china/science/article/3272844/chinese-police-ai-algorithm-tracks-bitcoin-money-laundering-90-accuracy].
• Computational Costs: The algorithms are computationally intensive, often requiring batch processing rather than providing real-time, streaming analysis of the blockchain [Source: https://www.scmp.com/news/china/science/article/3272844/chinese-police-ai-algorithm-tracks-bitcoin-money-laundering-90-accuracy].
• Black Box Problem: Despite the use of LLMs for "interpretability," the underlying deep learning models can be opaque, making it difficult to use their outputs as sole evidence in transparent court cases [Source: https://www.mdpi.com/journal/ai/special_issues/AI_Cryptocurrency].

In summary, China's AI tool is highly effective at identifying patterns in the Bitcoin ledger, but its real-world accuracy is heavily dependent on the ability of authorities to bridge the gap between on-chain data and off-chain identity through exchange cooperation.