Key Highlights
- Chainalysis says it does not use machine learning to identify wallet segments or establish wallet-control relationships.
- The firm classifies wallet clustering as Tier 1 structural intelligence, requiring deterministic and auditable methods.
- AI is used for lead generation, anomaly detection and pattern recognition, with outputs requiring further validation.
Chainalysis, a blockchain analytics firm, has warned that machine learning should not be treated as definitive evidence when analyzing blockchain activity, saying predictive models can introduce errors if their outputs are accepted without further verification.
In a report published August 14, the blockchain analytics firm said machine learning can help process large amounts of on-chain data, detect patterns and support investigations. However, it said the technology has limitations when used to make foundational claims about which wallets are controlled by the same entity.
Chainalysis draws line on wallet clustering
Chainalysis said it does not use machine learning to identify wallet segments, which involve determining whether different blockchain addresses are controlled by the same entity.
The company classifies these as “Tier 1” structural intelligence claims and said they must be deterministic, reproducible, and auditable.
Chainalysis said predictive models do not meet this standard because their conclusions are based on learned patterns from training data rather than fixed, independently reviewable rules.
The firm said even a highly accurate predictive model would not meet its structural soundness standard for these types of claims because investigators need to be able to reproduce and examine how the conclusion was reached.
How AI still used for investigative leads
The company said machine learning remains part of its work in areas where the output is treated as a signal rather than a definitive conclusion.
These include lead generation, anomaly detection, pattern recognition and evidence-based category assessments.
Chainalysis said such outputs can help investigators decide where to look more closely, provided they are treated as probabilistic assessments and undergo additional validation.
The company also said it uses AI and machine learning in its Alterya scam-detection tool, which analyzes web data, chat messages and blockchain activity to identify emerging scams.
Court challenges could put methods under scrutiny
Chainalysis also raised the issue of how blockchain analytics may be evaluated in court.
Under the U.S. Daubert standard, courts can examine whether expert evidence is based on a testable methodology, has undergone peer review, has a known error rate, and is generally accepted within its field.
The company cited the 2024 United States v. Sterlingov case, where the defense challenged Chainalysis’ clustering methodology.
According to Chainalysis, the court found its methodology sufficiently sound after examining how the clusters were constructed and whether the reasoning could be independently verified.
The company noted that the ruling addressed its specific methodology and did not establish blockchain analytics as a whole as automatically reliable or admissible.
Chainalysis said machine-learning-heavy approaches could face additional scrutiny if providers cannot explain how a wallet cluster was created or why a model reached a particular conclusion.
Errors could affect investigations and compliance
The report also highlights the potential consequences of incorrect wallet connections.
For law enforcement, an inaccurate link between addresses could send an investigation toward the wrong individual or entity.
For exchanges and financial institutions, a false connection between a customer’s wallet and a sanctioned address could lead to account restrictions, frozen funds or suspicious activity reports.
Prosecutors could also face challenges if the underlying methodology cannot be clearly explained in court.
AI and Blockchain research continues
The discussion comes as the crypto industry continues to explore other applications connecting artificial intelligence with blockchain infrastructure.
Grayscale Research said in an August 11 report that agent payments, verifiable records and decentralized AI are among areas where public blockchains could potentially address emerging requirements from AI systems.
Bittensor is another example of the intersection between the two technologies, using decentralized subnets for different machine-learning applications.
Those developments focus on using blockchain infrastructure for AI-related applications, while Chainalysis’ report addresses the separate question of how AI itself should be used to analyze blockchain data.
Transparency remains central to blockchain intelligence
Chainalysis’ position is that machine learning can assist blockchain investigations without replacing methods that allow investigators to verify the underlying evidence.
The company said predictive tools are better suited to generating leads and identifying patterns, while claims about wallet control require methodologies that are transparent, reproducible and auditable.
The distinction could become increasingly relevant as investigators, exchanges and financial institutions use automated tools to process growing volumes of blockchain activity.
