Disentangling Deception and Hallucination Failures in LLMs

TL;DR

This work reframes LLM failures by separating knowledge existence ($K$) from behavior expression ($B$), distinguishing hallucination ($K=0,B=1$) from deception ($K=1,B=0$). Using a controlled entity-centric QA setup with four entity types, the authors construct four behavioral regimes and verify knowledge preservation under deception via jailbreak probing. They analyze internal representations with a bottleneck classifier and sparse autoencoders to reveal early, $K$-driven separability and sparse, entity-dependent $B$-signals, and demonstrate causal control of behavior through layer-local angular activation editing that can induce transitions between correct and deceptive outputs but not to hallucination. The findings argue that evaluation and control of LLM reliability must operate at the mechanism level—addressing missing knowledge separately from regulated behavior—to improve safety and robustness in real-world deployments.

Abstract

Failures in large language models (LLMs) are often analyzed from a behavioral perspective, where incorrect outputs in factual question answering are commonly associated with missing knowledge. In this work, focusing on entity-based factual queries, we suggest that such a view may conflate different failure mechanisms, and propose an internal, mechanism-oriented perspective that separates Knowledge Existence from Behavior Expression. Under this formulation, hallucination and deception correspond to two qualitatively different failure modes that may appear similar at the output level but differ in their underlying mechanisms. To study this distinction, we construct a controlled environment for entity-centric factual questions in which knowledge is preserved while behavioral expression is selectively altered, enabling systematic analysis of four behavioral cases. We analyze these failure modes through representation separability, sparse interpretability, and inference-time activation steering.

Figures (9)

• Figure 1: External and Internal Views of LLM Outputs. We characterize model failures, including hallucination and deception, through two latent factors: Knowledge Existence and Behavioral Expression, which together provide an internal, mechanism-level account of how different failure modes arise.
• Figure 2: Construction of the controlled environment. Inputs are partitioned into knowledge-unkown and knowledge-unknown sets. The verified set is further split into $\mathcal{D}_{\mathrm{KC}}$, $\mathcal{D}_{\mathrm{KA}}$, and $\mathcal{D}_{\mathrm{KI}}$ via correct, evasive, or intentionally incorrect responses, followed by deceptive DPO to obtain $\mathcal{D}'$ under $L'$. Jailbreak probing verifies knowledge preservation, resulting in four behavioral cases.
• Figure 3: Overview of the representation analysis, interpretability, and intervention pipeline.Left (Sec. \ref{['sec: classification']}: Representational separability). We construct a bottleneck-based classifier on hidden representations of the derived model $L'$, jointly optimizing reconstruction and classification losses. The compressed bottleneck representations are used to assess the separability of the four behavioral cases $\mathcal{D}_{\mathrm{KC}}$, $\mathcal{D}_{\mathrm{KA}}$, $\mathcal{D}_{\mathrm{KI}}$, and $\mathcal{D}_{\mathrm{hal}}$. Middle (Sec. \ref{['sec: sae']}: Interpretability via sparse autoencoding). A post-hoc sparse autoencoder (SAE) is applied to the bottleneck representations to obtain sparse latent codes. This decomposition enables the analysis of shared and behavior-specific latent factors associated with different behavioral cases. Right (Sec. \ref{['sec: steering']}: Steering via angular activation editing). To causally test whether deceptive behaviors arise from suppressed behavior expression under fixed knowledge, we perform layer-local angular activation editing. Activations are rotated within a low-dimensional steering plane spanned by behavior-associated directions, enabling controlled transitions between deceptive and correct outputs without modifying model parameters or injecting external knowledge.
• Figure 4: Classification results under multi-class and binary settings.Left: Confusion matrix reporting classification accuracy (ACC) for a four-way classifier distinguishing the four behavioral cases. Right: Binary classification accuracy under different partitions corresponding to knowledge and behavior dimensions.
• Figure 5: Cross-layer separability analysis. We report the one-vs-rest AUC of the classifier as a function of the layer depth.