mlsanityData to 14 Oct 2026

Keep Calm and Avoid Harmful Content: Concept Alignment and Latent Manipulation Towards Safer Answers

arXiv 2510.12672 ↗First posted 14 Oct 2025cs.LGcited by 0

On the map: LLM safety and security → Safety alignment, refusal and guardrails

Abstract (arXiv)

Large Language Models are susceptible to jailbreak attacks that bypass built-in safety guardrails (e.g., by tricking the model with adversarial prompts). We propose Concept Alignment and Concept Manipulation CALM, an inference-time method that suppresses harmful concepts by modifying latent representations of the last layer of the model, without retraining. Leveraging concept whitening technique from Computer Vision combined with orthogonal projection, CALM removes unwanted latent directions associated with harmful content while preserving model performance. Experiments show that CALM reduces harmful outputs and outperforms baseline methods in most metrics, offering a lightweight approach to AI safety with no additional training data or model fine-tuning, while incurring only a small computational overhead at inference.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security. Builds on 32 of its 39 references that are in the map. First posted on arXiv 14 Oct 2025.

Builds on

32 of 39 references.

Closest ancestors · 4

Same direction, newest first

  1. 2024SCAR: Sparse Conditioned Autoencoders for Concept Detection and Steering in LLMs11 Nov
  2. 2024Refusal in Language Models Is Mediated by a Single Direction17 Jun
  3. 2024Improving Alignment and Robustness with Circuit Breakers6 Jun
  4. 2024Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications7 Feb

Other references · 28

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,720
  3. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  4. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,998
  5. 2023Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injectioncited by 1,443
  6. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  7. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  8. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
  9. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  10. 2023Prompt Injection attack against LLM-integrated Applicationscited by 605
  11. 2023GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Promptscited by 560
  12. 2023Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrationscited by 431
16 more
  1. 2023LEACE: Perfect linear concept erasure in closed formcited by 267
  2. 2023Pretraining Language Models with Human Preferencescited by 247
  3. 2020Concept Whitening for Interpretable Image Recognitioncited by 225
  4. 2023Red-Teaming Large Language Models using Chain of Utterances for Safety-Alignmentcited by 223
  5. 2023Scalable and Transferable Black-Box Jailbreaks for Language Models via Persona Modulationcited by 214
  6. 2021Whitening Sentence Representations for Better Semantics and Faster Retrievalcited by 202
  7. 2024Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMscited by 173
  8. 2024Adaptive Activation Steering: A Tuning-Free LLM Truthfulness Improvement Method for Diverse Hallucinations Categoriescited by 77
  9. 2024Spectral Editing of Activations for Large Language Model Alignmentcited by 56
  10. 2024Revisiting Jailbreaking for Large Language Models: A Representation Engineering Perspectivecited by 48
  11. 2024A Language Model's Guide Through Latent Spacecited by 43
  12. 2024Model Editing as a Robust and Denoised variant of DPO: A Case Study on Toxicitycited by 35
  13. 2025Controlling Large Language Models Through Concept Activation Vectorscited by 28
  14. 2019BERT has a Moral Compass: Improvements of ethical and moral values of machinescited by 21
  15. 2023Whitening-based Contrastive Learning of Sentence Embeddingscited by 21
  16. 2024Whitening Not Recommended for Classification Tasks in LLMscited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Safety Alignment Can Be Not Superficial With Explicit Safety Signals
  2. 2026Truth Lies Deep: Countering Semantic Camouflage via Latent Intent Verification
  3. 2025Vulnerability Mitigation for Safety-Aligned Language Models via Debiasing
  4. 2025Concept Concentration for Faithful Representation Intervention
  5. 2025SafeMERGE: Preserving Safety Alignment in Fine-Tuned Large Language Models via Selective Layer-Wise Model Merging
  6. 2025Probing the Robustness of Large Language Models Safety to Latent Perturbations
  7. 2026SafeAnchor: Preventing Cumulative Safety Erosion in Continual Domain Adaptation of Large Language Models
  8. 2026Beyond Routine Compliance: Cunning Data Cultivates Safety Vigilance in Large Language Models