CausalGaze: Unveiling Hallucinations via Counterfactual Graph Intervention in Large Language Models

2026-04-13

AI Summary

This work addresses the challenge of hallucinations in large language models (LLMs), which hinder their reliable deployment in high-stakes scenarios. The authors propose a novel approach grounded in structural causal models (SCMs), representing the internal states of LLMs as dynamic causal graphs. For the first time, they introduce an active counterfactual intervention mechanism that shifts hallucination detection from passive observation to causally driven reasoning, effectively disentangling genuine causal inference pathways from spurious noise. This method substantially improves both detection accuracy and interpretability, demonstrating consistent gains across four benchmark datasets and three prominent LLMs. Notably, it achieves an absolute AUROC improvement of over 5.2% on TruthfulQA compared to the current state-of-the-art.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Reasoning under Uncertainty: Causality

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Despite the groundbreaking advancements made by large language models (LLMs), hallucination remains a critical bottleneck for their deployment in high-stakes domains. Existing classification-based methods mainly rely on static and passive signals from internal states, which often captures the noise and spurious correlations, while overlooking the underlying causal mechanisms. To address this limitation, we shift the paradigm from passive observation to active intervention by introducing CausalGaze, a novel hallucination detection framework based on structural causal models (SCMs). CausalGaze models LLMs' internal states as dynamic causal graphs and employs counterfactual interventions to disentangle causal reasoning paths from incidental noise, thereby enhancing model interpretability. Extensive experiments across four datasets and three widely used LLMs demonstrate the effectiveness of CausalGaze, especially achieving over 5.2\% improvement in AUROC on the TruthfulQA dataset compared to state-of-the-art baselines.

Problem

• hallucination
• large language models
• causal mechanisms
• counterfactual intervention
• structural causal models

Innovation

• causal inference
• counterfactual intervention
• hallucination detection
• structural causal models
• large language models