mlsanityData to 14 Oct 2026

Steering the CensorShip: Uncovering Representation Vectors for LLM "Thought" Control

arXiv 2504.17130 ↗First posted 23 Apr 2025cs.CL cs.CR cs.CYcited by 14

On the map: LLM reasoning → Prompting and chain-of-thought reasoning

Abstract (arXiv)

Large language models (LLMs) have transformed the way we access information. These models are often tuned to refuse to comply with requests that are considered harmful and to produce responses that better align with the preferences of those who control the models. To understand how this "censorship" works. We use representation engineering techniques to study open-weights safety-tuned models. We present a method for finding a refusal--compliance vector that detects and controls the level of censorship in model outputs. We also analyze recent reasoning LLMs, distilled from DeepSeek-R1, and uncover an additional dimension of censorship through "thought suppression". We show a similar approach can be used to find a vector that suppresses the model's reasoning process, allowing us to remove censorship by applying the negative multiples of this vector. Our code is publicly available at: https://github.com/hannahxchen/llm-censorship-steering

Key facts. Cited by 14 papers in the map (data through 14 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 56 of its 71 references that are in the map. First posted on arXiv 23 Apr 2025.

Builds on

56 of 71 references.

Closest ancestors · 1

Same direction, newest first

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Models28 Jan

Other references · 55

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  5. 2023Qwen Technical Reportcited by 3,344
  6. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  7. 2022Constitutional AI: Harmlessness from AI Feedbackcited by 3,050
  8. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,651
  9. 2016Understanding intermediate layers using linear classifier probescited by 1,353
  10. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  11. 2023Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversationscited by 1,207
  12. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
38 more
  1. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  2. 2024Gemma: Open Models Based on Gemini Research and Technologycited by 916
  3. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Datasetcited by 860
  4. 2023AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Modelscited by 785
  5. 2023Whose Opinions Do Language Models Reflect?cited by 773
  6. 2024Yi: Open Foundation Models by 01.AIcited by 719
  7. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 716
  8. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 712
  9. 2023GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Promptscited by 560
  10. 2024JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Modelscited by 536
  11. 2023XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Modelscited by 505
  12. 2024WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMscited by 496
  13. 2023Catastrophic Jailbreak of Open-source LLMs via Exploiting Generationcited by 451
  14. 2023Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrationscited by 431
  15. 2023Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructionscited by 387
  16. 2024A StrongREJECT for Empty Jailbreakscited by 385
  17. 2023JudgeLM: Fine-tuned Large Language Models are Scalable Judgescited by 367
  18. 2023NeMo Guardrails: A Toolkit for Controllable and Safe LLM Applications with Programmable Railscited by 365
  19. 2024Deliberative Alignment: Reasoning Enables Safer Language Modelscited by 308
  20. 2024SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusalcited by 240
  21. 2024JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attackscited by 230
  22. 2024Programming Refusal with Conditional Activation Steeringcited by 206
  23. 2024Imitate, Explore, and Self-Improve: A Reproduction Report on Slow-thinking Reasoning Systemscited by 164
  24. 2023MART: Improving LLM Safety with Multi-round Automatic Red-Teamingcited by 163
  25. 2025SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilitiescited by 136
  26. 2023Open Sesame! Universal Black Box Jailbreaking of Large Language Modelscited by 134
  27. 2022The Ghost in the Machine has an American accent: value conflict in GPT-3cited by 105
  28. 2025H-CoT: Hijacking the Chain-of-Thought Safety Reasoning Mechanism to Jailbreak Large Reasoning Models, Including OpenAI o1/o3, DeepSeek-R1, and Gemini 2.0 Flash Thinkingcited by 99
  29. 2025OverThink: Slowdown Attacks on Reasoning LLMscited by 96
  30. 2023Cognitive Overload: Jailbreaking Large Language Models with Overloaded Logical Thinkingcited by 81
  31. 2024BadChain: Backdoor Chain-of-Thought Prompting for Large Language Modelscited by 80
  32. 2024Unintended Impacts of LLM Alignment on Global Representationcited by 66
  33. 2025Reasoning-Augmented Conversation for Multi-Turn Jailbreak Attacks on Large Language Modelscited by 64
  34. 2024Against The Achilles' Heel: A Survey on Red Teaming for Generative Modelscited by 59
  35. 2023LLM Censorship: A Machine Learning Challenge or a Computer Security Problem?cited by 49
  36. 2024Surgical, Cheap, and Flexible: Mitigating False Refusal in Language Models via Single Vector Ablationcited by 48
  37. 2024Multi-property Steering of Large Language Models with Dynamic Activation Compositioncited by 39
  38. 2024AI Risk Categorization Decoded (AIR 2024): From Government Regulations to Corporate Policiescited by 37

50 most cited of 55 All 55 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 7; 2026: 7.

14 papers.

Sort:

By area:

  1. 2025A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Modelscited by 25
  2. 2025SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignmentcited by 24
  3. 2025Beyond Prompt Engineering: Robust Behavior Control in LLMs via Steering Target Atomscited by 24
  4. 2025Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Trainingcited by 16
  5. 2025Comparative Analysis of LLM Abliteration Methods: A Cross-Architecture Evaluationcited by 9
  6. 2025Refusal Steering: Fine-grained Control over LLM Refusal Behaviour for Sensitive Topicscited by 6
  7. 2026When Safety Becomes a Vulnerability: Exploiting LLM Alignment Homogeneity for Transferable Blocking in RAGcited by 4
  8. 2026How Alignment Routes: Localizing, Scaling, and Controlling Policy Circuits in Language Modelscited by 3
  9. 2025Steering Information Utility in Key-Value Memory for Language Model Post-Trainingcited by 2
  10. 2026Do LLMs Know What Is Private Internally? Probing and Steering Contextual Privacy Norms in Large Language Model Representationscited by 2
  11. 2026AntiPaSTO: Self-Supervised Honesty Steering via Anti-Parallel Representationscited by 0
  12. 2026White-Box Sensitivity Auditing with Steering Vectorscited by 0
2 more
  1. 2026Defeat Devices in AI Systemscited by 0
  2. 2026Detecting Hidden Behaviors in LLMs via Activation-matched Finetuningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025SafeSteer: Interpretable Safety Steering with Refusal-Evasion in LLMs
  2. 2025Understanding and Mitigating Over-refusal for Large Language Models via Representation Intervention
  3. 2026What the Model Cannot See: Covert and Provable Text Invisibility in LLMs
  4. 2025Information Suppression in Large Language Models: Auditing, Quantifying, and Characterizing Censorship in DeepSeek
  5. 2025Watch the Weights: Unsupervised monitoring and control of fine-tuned LLMs
  6. 2024Jailbreak Antidote: Runtime Safety-Utility Balance via Sparse Representation Adjustment in Large Language Models
  7. 2025RepIt: Steering Language Models with Concept-Specific Refusal Vectors
  8. 2024On Prompt-Driven Safeguarding for Large Language Models