HalluGraph: Auditable Hallucination Detection for Legal RAG Systems via Knowledge Graph Alignment

TL;DR

HalluGraph is a graph-based framework that detects hallucinations in legal AI systems by aligning knowledge graphs from context, query, and response, providing interpretable metrics for accountability.

Contribution

It introduces a novel graph-theoretic approach for verifiable hallucination detection in legal RAG systems, improving transparency and reliability over existing semantic similarity methods.

Findings

Achieves near-perfect discrimination on structured control documents.

Attains AUC = 0.979 on legal hallucination detection.

Maintains robust performance (AUC ≈ 0.89) on challenging legal tasks.

Abstract

Legal AI systems powered by retrieval-augmented generation (RAG) face a critical accountability challenge: when an AI assistant cites case law, statutes, or contractual clauses, practitioners need verifiable guarantees that generated text faithfully represents source documents. Existing hallucination detectors rely on semantic similarity metrics that tolerate entity substitutions, a dangerous failure mode when confusing parties, dates, or legal provisions can have material consequences. We introduce HalluGraph, a graph-theoretic framework that quantifies hallucinations through structural alignment between knowledge graphs extracted from context, query, and response. Our approach produces bounded, interpretable metrics decomposed into \textit{Entity Grounding} (EG), measuring whether entities in the response appear in source documents, and \textit{Relation Preservation} (RP), verifying that asserted relationships are supported by context. On structured control documents, HalluGraph achieves near-perfect discrimination ($>$400 words, $>$20 entities), HalluGraph achieves $AUC = 0.979$, while maintaining robust performance ($AUC \approx 0.89$) on challenging generative legal task, consistently outperforming semantic similarity baselines. The framework provides the transparency and traceability required for high-stakes legal applications, enabling full audit trails from generated assertions back to source passages.