Steer-to-Detect: Probing Hidden Representations for Detection of LLM-Generated Texts
Abstract
The rapid advancement of large language models (LLMs) has made machine-generated text increasingly difficult to distinguish from human-written text. While recent studies explore leveraging internal representations of language models to uncover deeper detection signals, these raw features often exhibit substantial overlap between classes, limiting their discriminative power. To address this challenge, we propose Steer-to-Detect (\texttt{S2D}), a two-stage framework for detecting LLM-generated text. In the first stage, \texttt{S2D} learns a steering vector that is injected into the hidden states of a frozen observer LLM, producing representations with improved class separability. In the second stage, detection is performed via a hypothesis testing procedure based on the steered representations. We establish finite-sample, high-probability guarantees for Type I and Type II errors, providing a theoretical characterization of the procedure. Empirically, \texttt{S2D} achieves strong and consistent performance across a range of settings, including out-of-distribution scenarios and adversarial perturbations.
Related
- Counterfactual Probing for Parallel Unmasking with Hidden Forest Structure
- OUT-OF-DISTRIBUTION AI-GENERATED IMAGE DETECTION WITH ZERO-PHASE DECONVOLUTION FOR RESIDUAL WHITENING
- Robust Detection of LLM-Generated Text under Contamination
- Evaluating Xception with Multitask Self-Supervised Learning for AI-Generated Image and Deepfake Video Detection
- SHOCKING! College professor reveals hidden secrets of news headlines. Can you guess? Examining the efficacy of clickbait detection training and news literacy
BibTeX
@misc{liang2026steertodetect,
title = {{Steer-to-Detect: Probing Hidden Representations for Detection of LLM-Generated Texts}},
author = {Luxu Liang and Xiang Li},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2605.12890},
eprint = {2605.12890},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2605.12890},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Steer-to-Detect: Probing Hidden Representations for Detection of LLM-Generated Texts
AU - Luxu Liang
AU - Xiang Li
PY - 2026
DA - 2026/09/30
PB - arXiv
DO - 10.48550/arxiv.2605.12890
UR - https://arxiv.org/abs/2605.12890
AB - The rapid advancement of large language models (LLMs) has made machine-generated text increasingly difficult to distinguish from human-written text. While recent studies explore leveraging internal representations of language models to uncover deeper detection signals, these raw features often exhibit substantial overlap between classes, limiting their discriminative power. To address this challenge, we propose Steer-to-Detect (\texttt{S2D}), a two-stage framework for detecting LLM-generated text. In the first stage, \texttt{S2D} learns a steering vector that is injected into the hidden states of a frozen observer LLM, producing representations with improved class separability. In the second stage, detection is performed via a hypothesis testing procedure based on the steered representations. We establish finite-sample, high-probability guarantees for Type I and Type II errors, providing a theoretical characterization of the procedure. Empirically, \texttt{S2D} achieves strong and consistent performance across a range of settings, including out-of-distribution scenarios and adversarial perturbations.
ER -