TL;DR: Blockchain analytics still relies on the overloaded idea of a “cluster,” but the paper argues that address grouping, entity attribution, and operator determination are distinct analytical operations with different evidence standards and failure modes, according to Chainalysis. That separation matters because compliance and investigation decisions are only as reliable as the evidentiary boundary behind them.
At a glance
What this is: This paper argues that blockchain address clustering is not one analytical act but three distinct operations that require different evidence standards and confidence handling.
Why it matters: It matters because compliance, investigations, and customer-risk decisions become unreliable when teams treat structurally reproducible facts and attribution claims as if they carry the same evidentiary weight.
Context
Blockchain analytics is often used to support compliance screening, investigations, sanctions decisions, and asset tracing, so the evidence model behind it has operational consequences. The problem is not whether clustering is useful, but whether teams are clear about what kind of claim they are making when they group addresses or assign an entity.
In this paper, Chainalysis argues that the industry has overloaded a single term to cover different analytical acts with different failure modes. That creates a governance gap for compliance teams, because the question is not only whether a model works, but whether its output is being interpreted as structure, attribution, or operator behaviour.
Key questions
Q: How should compliance teams separate structural blockchain findings from attribution claims?
A: Treat structural findings as reproducible analytical outputs and attribution as a higher-uncertainty intelligence judgement. The practical test is whether another analyst can reach the same result from the same inputs. If not, the claim belongs in a different evidentiary class and should carry confidence language before it enters a decision process.
Q: Why do blockchain analytics decisions become risky when clustering is treated as one thing?
A: Because one label can hide three different kinds of inference with different error rates and consequences. Address grouping may be repeatable, but entity attribution and operator determination depend on weaker evidence and more interpretation. When teams collapse those layers, they risk overclaiming certainty in compliance, investigations, and legal escalation.
Q: What are the signs that an analytics model is crossing from lead generation into proof?
A: The warning signs are absolute language, missing source characterisation, and outputs that are consumed as final answers rather than hypotheses. If a model result is being used to justify a compliance action without an evidentiary label, confidence level, or reviewer challenge, it has crossed the boundary into unsupported proof.
Q: What should teams do when vendor analytics uses vague clustering terminology?
A: Require the vendor to map each output to a named claim type, explain the evidence standard behind it, and state where machine learning is only generating leads. If the terminology remains ambiguous, the safe assumption is that the product is blending structure and attribution in ways that can mislead downstream users.
Technical breakdown
Why clustering is not a single analytic claim
A blockchain address cluster can mean more than one thing in practice. Address grouping is a structural claim about likely wallet linkage, entity attribution is a higher-level claim about who controls the activity, and operator determination is a behavioural inference about who is acting at a given moment. Those are not interchangeable because they rest on different kinds of evidence and produce different levels of certainty. Mixing them creates a false impression that every cluster output is equally authoritative, when some outputs are best treated as deterministic structure and others as probabilistic intelligence.
Practical implication: Practitioners should separate structural outputs from attribution outputs in workflows, reports, and review thresholds.
Two-tier evidence changes how blockchain intelligence should be used
A two-tier evidence model gives the discipline a cleaner boundary. Structural claims should be deterministic and reproducible, meaning another analyst should be able to reach the same result from the same observable inputs. Attribution and intelligence claims, by contrast, depend on explicit source characterization, confidence levels, and context that can support or weaken the inference. This distinction matters because compliance and legal teams often need to know whether a result can stand as a verifiable analytical fact or whether it is a reasoned judgment that still carries uncertainty.
Practical implication: Teams should require different review standards for reproducible structure than for source-based attribution.
Where machine learning fits and where it does not
Machine learning can help surface leads, rank possibilities, or identify patterns that warrant review, but it should not be allowed to collapse evidentiary categories. If a model proposes a likely linkage, that is not the same as proving an address group or establishing operator identity. The ontology proposed in the paper is trying to prevent exactly that drift by making the evidence standard visible at the point of analysis. Without that boundary, automated scoring can quietly move from assistance into unsupported certainty.
Practical implication: Use machine learning for lead generation, but keep final structural and attribution claims under explicit evidentiary controls.
Threat narrative
Attacker objective: The objective is not a classic external intrusion but analytical ambiguity that can distort downstream compliance, tracing, and enforcement decisions.
- Entry occurs when investigators or compliance teams start from an overloaded cluster concept that hides distinct analytical operations behind one label.
- Escalation happens when attribution, entity grouping, and operator determination are treated as equivalent outputs even though they carry different evidence standards.
- Impact is misclassification in compliance, investigation, or legal decision-making because the evidentiary basis for the claim was never made explicit.
Breaches seen in the wild
- Shai Hulud npm malware campaign: Shai Hulud campaign: npm malware exposed secrets on GitHub.
Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.
NHI Mgmt Group analysis
Formal evidence boundaries are now a governance requirement, not a taxonomy exercise. Once blockchain analytics informs compliance or investigative action, the question is no longer whether a label is convenient, but whether the evidence behind it is defensible. Chainalysis is pointing to a larger control problem in security analytics: teams routinely operationalise outputs before defining what standard of proof those outputs actually meet. The practitioner conclusion is to classify claim types before they are consumed by policy, casework, or legal review.
The industry needs a shared vocabulary because ambiguity creates audit risk. When one term carries structural linkage, attribution, and behavioural inference, teams cannot consistently explain why a decision was made or how confident they were at the time. That creates avoidable friction across compliance, investigations, and vendor evaluation. The practitioner conclusion is to adopt language that distinguishes deterministic findings from intelligence judgments.
Evidence-tier separation: the decisive control is not better clustering, but clearer claim classification. This paper matters because it reframes the problem from analytic capability to governance of analytic certainty. In identity-adjacent investigations, the same principle applies whenever tools infer relationships from partial signals. The practitioner conclusion is to demand a claim model that makes confidence, source type, and reproducibility explicit.
Machine learning should support discovery, not overwrite evidentiary standards. Predictive systems can accelerate triage, but they cannot make an inference more certain than the underlying evidence allows. That distinction matters in both blockchain intelligence and broader fraud or identity investigations, where a lead is often mistaken for proof. The practitioner conclusion is to keep human review and evidentiary labelling in the decision path.
This is a useful model for any security domain that confuses signal generation with proof. The article is about blockchain analytics, but the governance lesson extends to identity verification, fraud operations, and NHI monitoring where classifiers can surface patterns faster than they can justify conclusions. The practitioner conclusion is to treat confidence metadata as part of the control, not an optional annotation.
What this signals
Claim discipline will matter more as blockchain analytics moves deeper into compliance workflows. The key programme risk is not bad data alone, but analysts and systems turning different levels of certainty into a single operational answer. Teams that do not distinguish reproducible structure from intelligence inference will struggle to explain decisions later.
Evidence-tier separation: once that distinction is formalised, it becomes easier to govern analyst judgment, vendor evaluation, and case escalation. That same discipline is relevant in fraud and identity operations, where signal generation is often mistaken for proof.
For practitioners
- Define evidence classes for every analytical claim Separate structural, attribution, and behavioural claims in your reporting so each one carries a distinct evidentiary standard and review path.
- Require confidence labels on attribution outputs Do not let intelligence-based conclusions enter casework or compliance decisions without explicit source characterization and confidence language.
- Keep machine learning in the lead-generation layer Use predictive models to surface candidates for review, but reserve final classification and operator attribution for governed analyst decisions.
- Standardise terms across legal, compliance, and investigations Publish a common vocabulary that distinguishes address grouping from entity attribution and operator determination before teams rely on vendor outputs.
Key takeaways
- Blockchain clustering is not one claim but three, and each one needs its own evidence standard before it is used operationally.
- The main governance risk is overconfidence, because structural linkage and attribution do not carry the same level of certainty.
- Teams should separate machine-generated leads from final analytical proof if they want defensible compliance and investigation outcomes.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
NIST CSF 2.0, NIST AI RMF, NIST SP 800-53 Rev 5 and CIS Controls v8 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | ID.AM-01 — Physical Devices and Systems Inventory | The article is about making analytical outputs inventoryable and distinguishable before they drive decisions. |
| GV.RM-01 — Risk Management Strategy | The paper is fundamentally about governing uncertainty in compliance and investigation decisions. | |
| Recommendation — Map analytical outputs to distinct claim types so downstream teams know exactly what each result represents. Define a risk strategy that distinguishes reproducible findings from probabilistic intelligence claims. | ||
| NIST AI RMF | GOVERN — AI Governance and Accountability | Machine learning is discussed as a lead-generation tool that must be governed rather than trusted as proof. |
| Recommendation — Set governance rules that keep model outputs labelled as leads until evidence supports stronger claims. | ||
| NIST SP 800-53 Rev 5 | AU-6 — Audit Review, Analysis, and Reporting | The article emphasizes reportability and accountability for evidentiary standards in analytics. |
| Recommendation — Standardise reporting so analysts record evidence basis, confidence, and review status for each claim. | ||
| CIS Controls v8 | CIS-8 — Audit Log Management | The governance problem is traceability of how analytical conclusions were formed and consumed. |
| Recommendation — Preserve decision trails that show when a result was a structural fact versus an attributed inference. | ||
Key terms
- Address Grouping: Address grouping is the process of linking multiple blockchain addresses together based on observed technical or behavioural patterns. It produces a structural analytic claim, not identity proof, and should be treated as a lead unless the method is fully deterministic and reproducible.
- Agent Attribution: Agent attribution is the ability to tie each request, action, retry, and cost event back to a specific non-human actor. It matters because shared credentials and anonymous automation hide misuse, make revocation harder, and prevent security teams from understanding which identity actually performed a task.
- Operator Determination: Operator determination is the inference that a specific actor controlled a wallet, address cluster, or transaction sequence at a given time. It is the weakest of the three claims described in the article unless supported by strong corroborating evidence, so it should never be mistaken for direct identity verification.
- Two-tier Evidence Model: A two-tier evidence model separates deterministic structural findings from intelligence-driven attribution. The purpose is to make confidence, reproducibility, and source quality visible so that compliance, legal, and investigative decisions are made on the right kind of claim.
Deepen your knowledge
The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and secrets management. It helps practitioners connect identity control design to the wider security decisions their programme depends on.
Published by the NHIMG editorial team on July 10, 2026.
Updated on October 11, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org