Explainable AML Triage with LLMs: Evidence Retrieval and Counterfactual Checks

2026-04-24 · Source: cs.AI updates on arXiv.org · Field: Finance & Economics — FinTech & Digital Financial Services, Banking & Financial Services · Depth: Expert, quick

Summary

A new explainable anti-money laundering (AML) triage framework leverages large language models (LLMs) to process high volumes of alerts under strict audit and governance constraints. The framework addresses risks like hallucinations and weak provenance in regulated workflows by treating triage as an evidence-constrained decision process. It integrates retrieval-augmented evidence bundling from diverse sources, a structured LLM output contract requiring explicit citations and evidence separation, and counterfactual checks to validate decision coherence under minimal perturbations. Evaluated on public synthetic AML benchmarks and simulators, the method achieved a PR-AUC of 0.75 and an Escalate F1 of 0.62, outperforming rules, tabular/graph ML baselines, and LLM-only/RAG-only variants. It also demonstrated strong provenance and faithfulness metrics, including a citation validity of 0.98, evidence support of 0.88, and counterfactual faithfulness of 0.76.

Key takeaway

For research scientists developing AI solutions in regulated industries like finance, this framework demonstrates how to build LLM-powered systems that meet stringent compliance requirements. You should integrate evidence retrieval, structured output with explicit citations, and counterfactual validation to mitigate hallucination risks and ensure auditability, thereby enabling practical decision support without sacrificing traceability and defensibility.

Key insights

Governed LLM systems can provide verifiable decision support for AML triage while meeting compliance requirements.

Principles

• Evidence grounding improves auditability.
• Counterfactual validation increases robustness.

Method

Combine retrieval-augmented evidence bundling, structured LLM output with citations, and counterfactual checks to ensure explainable, robust AML triage.

In practice

• Use structured LLM output contracts.
• Implement counterfactual checks for validation.

Topics

• Anti-Money Laundering Triage
• Large Language Models
• Explainable AI
• Evidence Retrieval
• Counterfactual Explanations

Best for: Research Scientist, AI Scientist, Director of AI/ML, Legal Professional

Related on AIssential

• A Definition of Good Explanations and the Challenges Explaining LLM Outputs — Good AI explanations combine counterfactuals with the interlocutor's prior beliefs to address LLM explainability challenges.
• Agentic Search for Counterfactual Recourse under Fixed LLM Budgets — Efficiently generating a diverse set of oracle-validated counterfactuals under fixed LLM call budgets is crucial for actionable recourse.
• LLMs Can’t Provide Faithful Explanations Needed for AI Accountability — LLMs struggle to provide faithful explanations, hindering accountability in high-stakes AI applications.
• TRIAGE: Dialectical Reasoning for Explainable Risk Prediction on Irregularly Sampled Medical Time Series with LLMs — TRIAGE uses dialectical LLM reasoning to provide calibrated, continuous risk scores and interpretable rationales for medical time series.
• Hallucinations in LLMs: A Deep Technical Dive into Causes, Detection, and Mitigation — LLM hallucinations are inherent to their probabilistic nature, driven by training objectives that prioritize fluency over factual accuracy.
• Decoupled Smart Contract Audits: Lightweight LLM Framework via Distillation and Aggregation — Lightweight LLMs can achieve high accuracy in smart contract auditing by decoupling tasks and employing distillation and aggregation strategies.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.