Saved in:
Bibliographic Details
Main Authors: Park, Ji-jun, Choi, Soo-joon, Jeong, Jiwon, Yoon, Taeyang, Lee, Ju-Wan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.28825
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917540812292096
author Park, Ji-jun
Choi, Soo-joon
Jeong, Jiwon
Yoon, Taeyang
Lee, Ju-Wan
author_facet Park, Ji-jun
Choi, Soo-joon
Jeong, Jiwon
Yoon, Taeyang
Lee, Ju-Wan
contents Large language models (LLMs) frequently encode factual and reasoning knowledge in their internal representations that is not faithfully reflected in their surface-level outputs -- a phenomenon known as \emph{latent knowledge}. Existing approaches to eliciting latent knowledge, such as Contrastive Consistency Search (CCS), rely on contrastive activation patterns and struggle with complex multi-step reasoning tasks, while mechanistic interpretability tools have primarily been used to \emph{understand} model behavior rather than to \emph{extract} hidden knowledge. We present \textbf{MechELK}, a unified three-stage framework that bridges mechanistic interpretability and latent knowledge elicitation. MechELK operates through: (1) \textbf{Locate} -- using Sparse Autoencoder (SAE) feature analysis and activation patching to identify knowledge-bearing representations; (2) \textbf{Verify} -- employing causal probing to distinguish genuine latent knowledge from spurious correlations; and (3) \textbf{Elicit} -- applying representation engineering to surface hidden knowledge without modifying model weights. Evaluated on TruthfulQA, a curated Deceptive Alignment benchmark, and the Quirky LM dataset, MechELK achieves an average elicitation accuracy of 84.7\%, outperforming CCS by 6.2\% and direct linear probing by 9.1\%. Crucially, MechELK successfully identifies latent knowledge in 78.3\% of cases where the model's surface output is incorrect or evasive, demonstrating its utility for AI safety applications including deceptive alignment detection.
format Preprint
id arxiv_https___arxiv_org_abs_2605_28825
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MechELK: A Mechanistic Interpretability Framework for Eliciting Latent Knowledge in Large Language Models
Park, Ji-jun
Choi, Soo-joon
Jeong, Jiwon
Yoon, Taeyang
Lee, Ju-Wan
Computation and Language
Large language models (LLMs) frequently encode factual and reasoning knowledge in their internal representations that is not faithfully reflected in their surface-level outputs -- a phenomenon known as \emph{latent knowledge}. Existing approaches to eliciting latent knowledge, such as Contrastive Consistency Search (CCS), rely on contrastive activation patterns and struggle with complex multi-step reasoning tasks, while mechanistic interpretability tools have primarily been used to \emph{understand} model behavior rather than to \emph{extract} hidden knowledge. We present \textbf{MechELK}, a unified three-stage framework that bridges mechanistic interpretability and latent knowledge elicitation. MechELK operates through: (1) \textbf{Locate} -- using Sparse Autoencoder (SAE) feature analysis and activation patching to identify knowledge-bearing representations; (2) \textbf{Verify} -- employing causal probing to distinguish genuine latent knowledge from spurious correlations; and (3) \textbf{Elicit} -- applying representation engineering to surface hidden knowledge without modifying model weights. Evaluated on TruthfulQA, a curated Deceptive Alignment benchmark, and the Quirky LM dataset, MechELK achieves an average elicitation accuracy of 84.7\%, outperforming CCS by 6.2\% and direct linear probing by 9.1\%. Crucially, MechELK successfully identifies latent knowledge in 78.3\% of cases where the model's surface output is incorrect or evasive, demonstrating its utility for AI safety applications including deceptive alignment detection.
title MechELK: A Mechanistic Interpretability Framework for Eliciting Latent Knowledge in Large Language Models
topic Computation and Language
url https://arxiv.org/abs/2605.28825