mlsanityData to 14 Oct 2026

Steer-to-Detect: Probing Hidden Representations for Detection of LLM-Generated Texts

arXiv 2605.12890 ↗First posted 13 May 2026stat.AP cs.LGcited by 0

On the map: Mechanistic interpretability → Activation steering methods

Abstract (arXiv)

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 (textttS2D), a two-stage framework for detecting LLM-generated text. In the first stage, textttS2D 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, textttS2D achieves strong and consistent performance across a range of settings, including out-of-distribution scenarios and adversarial perturbations.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Activation steering methods, one of 13 directions in Mechanistic interpretability. Builds on 56 of its 88 references that are in the map. First posted on arXiv 13 May 2026.

Builds on

56 of 88 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025Steering Evaluation-Aware Language Models to Act Like They Are Deployed23 Oct
  2. 2025Activation Steering with a Feedback Controller5 Oct
  3. 2025A Unified Understanding and Evaluation of Steering Methods4 Feb
  4. 2023Steering Llama 2 via Contrastive Activation Addition9 Dec
  5. 2023Steering Language Models With Activation Engineering20 Aug

Other references · 51

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  3. 2025Qwen3 Technical Reportcited by 8,820
  4. 2015Character-level Convolutional Networks for Text Classificationcited by 2,330
  5. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,186
  6. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  7. 2018Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarizationcited by 1,317
  8. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
  9. 2018Hierarchical Neural Story Generationcited by 1,193
  10. 2023PaLM 2 Technical Reportcited by 1,162
  11. 2023DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvaturecited by 726
  12. 2023A Watermark for Large Language Modelscited by 725
38 more
  1. 2019Release Strategies and the Social Impacts of Language Modelscited by 513
  2. 2019GLTR: Statistical Detection and Visualization of Generated Textcited by 496
  3. 2025Layer by Layer: Uncovering Hidden Representations in Language Modelscited by 311
  4. 2023Fast-DetectGPT: Efficient Zero-Shot Detection of Machine-Generated Text via Conditional Probability Curvaturecited by 287
  5. 2024Spotting LLMs With Binoculars: Zero-Shot Detection of Machine-Generated Textcited by 263
  6. 2019Automatic Detection of Generated Text is Easiest when Humans are Fooledcited by 261
  7. 2024LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinationscited by 244
  8. 2024Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Modelscited by 235
  9. 2023DetectLLM: Leveraging Log Rank Information for Zero-Shot Detection of Machine-Generated Textcited by 210
  10. 2023Instructions as Backdoors: Backdoor Vulnerabilities of Instruction Tuning for Large Language Modelscited by 110
  11. 2020Analyzing Individual Neurons in Pre-trained Language Modelscited by 97
  12. 2023Neurons in Large Language Models: Dead, N-gram, Positionalcited by 88
  13. 2023ChatGPT or Human? Detect and Explain. Explaining Decisions of Machine Learning Model for Detecting Short ChatGPT-generated Textcited by 85
  14. 2024Delving into LLM-assisted writing in biomedical publications through excess vocabularycited by 78
  15. 2022Do Language Models Plagiarize?cited by 73
  16. 2024(A)I Am Not a Lawyer, But...: Engaging Legal Experts towards Responsible LLM Policies for Legal Advicecited by 63
  17. 2024A Statistical Framework of Watermarks for Large Language Models: Pivot, Detection Efficiency and Optimal Rulescited by 57
  18. 2024Raidar: geneRative AI Detection viA Rewritingcited by 52
  19. 2025Steer LLM Latents for Hallucination Detectioncited by 50
  20. 2025AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraintcited by 49
  21. 2024DetectRL: Benchmarking LLM-Generated Text Detection in Real-World Scenarioscited by 38
  22. 2024Training-free LLM-generated Text Detection by Mining Token Probability Sequencescited by 35
  23. 2025Toward universal steering and monitoring of AI modelscited by 30
  24. 2025To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Modelscited by 29
  25. 2025RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patternscited by 24
  26. 2024Imitate Before Detect: Aligning Machine Stylistic Preference for Machine-Revised Text Detectioncited by 23
  27. 2025Spotlight Your Instructions: Instruction-following with Dynamic Attention Steeringcited by 21
  28. 2025AdaDetectGPT: Adaptive Detection of LLM-Generated Text with Statistical Guaranteescited by 21
  29. 2024Evaluation of LLM Vulnerabilities to Being Misused for Personalized Disinformation Generationcited by 20
  30. 2025LLM-Friendly Knowledge Representation for Customer Supportcited by 12
  31. 2026Detecting LLM-Generated Text with Performance Guaranteescited by 10
  32. 2026Learn-to-Distance: Distance Learning for Detecting LLM-Generated Textcited by 8
  33. 2026Efficient and accurate steering of Large Language Models through attention-guided feature learningcited by 8
  34. 2025SpecDetect: Simple, Fast, and Training-Free Detection of LLM-Generated Text via Spectral Analysiscited by 6
  35. 2025Hallucination reduction with CASAL: Contrastive Activation Steering For Amortized Learningcited by 6
  36. 2024Building Intelligence Identification System via Large Language Model Watermarking: A Survey and Beyondcited by 5
  37. 2025MGT-Prism: Enhancing Domain Generalization for Machine-Generated Text Detection via Spectral Alignmentcited by 5
  38. 2026Zero-Shot Detection of LLM-Generated Text via Implicit Reward Modelcited by 4

50 most cited of 51 All 51 (JSON)

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2023A Survey on LLM-Generated Text Detection: Necessity, Methods, and Future Directions
  2. 2026LLM Self-Recognition: Steering and Retrieving Activation Signatures
  3. 2026Detecting the Machine: A Comprehensive Benchmark of AI-Generated Text Detectors Across Architectures, Domains, and Adversarial Conditions
  4. 2026SV-Detect: AI-generated Text Detection with Steering Vectors
  5. 2025Feature-Level Insights into Artificial Text Detection with Sparse Autoencoders
  6. 2023From Text to Source: Results in Detecting Large Language Model-Generated Content
  7. 2025CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features
  8. 2026COLD-Steer: Steering Large Language Models via In-Context One-step Learning Dynamics