mlsanityData to 7 Oct 2026

The Geometry of Harmfulness in LLMs through Subconcept Probing

arXiv 2507.21141 ↗First posted 23 Jul 2025cs.AIcited by 7

Not on the map. Why?

Abstract (arXiv)

Recent advances in large language models (LLMs) have intensified the need to understand and reliably curb their harmful behaviours. We introduce a multidimensional framework for probing and steering harmful content in model internals. For each of 55 distinct harmfulness subconcepts (e.g., racial hate, employment scams, weapons), we learn a linear probe, yielding 55 interpretable directions in activation space. Collectively, these directions span a harmfulness subspace that we show is strikingly low-rank. We then test ablation of the entire subspace from model internals, as well as steering and ablation in the subspace's dominant direction. We find that dominant direction steering allows for near elimination of harmfulness with a low decrease in utility. Our findings advance the emerging view that concept subspaces provide a scalable lens on LLM behaviour and offer practical tools for the community to audit and harden future generations of language models.

Key facts. Cited by 7 papers in the map (data through 7 Oct 2026). Builds on 31 of its 42 references that are in the map. First posted on arXiv 23 Jul 2025.

Builds on

31 of 42 references.

References · 31

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  4. 2021On the Opportunities and Risks of Foundation Modelscited by 3,961
  5. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 2,914
  6. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,330
  7. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,259
  8. 2020RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Modelscited by 1,235
  9. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 960
  10. 2022Toy Models of Superpositioncited by 928
  11. 2023Steering Language Models With Activation Engineeringcited by 824
  12. 2023AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Modelscited by 738
19 more
  1. 2020Aligning AI With Shared Human Valuescited by 706
  2. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 679
  3. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 673
  4. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 412
  5. 2023Emergent Linear Representations in World Models of Self-Supervised Sequence Modelscited by 313
  6. 2023Large Language Model Unlearningcited by 288
  7. 2023Linear Representations of Sentiment in Large Language Modelscited by 157
  8. 2024Language Models are Homer Simpson! Safety Re-Alignment of Fine-tuned Language Models through Task Arithmeticcited by 105
  9. 2025The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independencecited by 104
  10. 2024On the Origins of Linear Representations in Large Language Modelscited by 76
  11. 2024pyvene: A Library for Understanding and Improving PyTorch Models via Interventionscited by 53
  12. 2024Enhancing LLM Safety via Constrained Direct Preference Optimizationcited by 47
  13. 2025SafeDPO: A Simple Approach to Direct Preference Optimization with Enhanced Safetycited by 37
  14. 2024Safety-Aware Fine-Tuning of Large Language Modelscited by 33
  15. 2025Interpretable Steering of Large Language Models with Feature Guided Activation Additionscited by 32
  16. 2024Mission Impossible: A Statistical Perspective on Jailbreaking LLMscited by 21
  17. 2025Refusal Behavior in Large Language Models: A Nonlinear Perspectivecited by 13
  18. 2025Detecting and Filtering Unsafe Training Data via Data Attribution with Denoised Representationcited by 9
  19. 2025From Directions to Cones: Exploring Multidimensional Representations of Propositional Facts in LLMscited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 4; 2026: 3.

7 papers.

  1. 2025Sycophancy Is Not One Thing: Causal Separation of Sycophantic Behaviors in LLMscited by 32
  2. 2026Before the Last Token: Diagnosing Final-Token Safety Probe Failurescited by 3
  3. 2025Any-Depth Alignment: Unlocking Innate Safety Alignment of LLMs to Any-Depthcited by 2
  4. 2026The Entanglement Wall: Activation-Space Probes as Risk Detectors, Not Context Adjudicatorscited by 2
  5. 2025NeuroBreak: Unveil Internal Jailbreak Mechanisms in Large Language Modelscited by 1
  6. 2025Estimating Dimensionality of Neural Representations from Finite Samplescited by 0
  7. 2026The Role of Fine-grained Harm Signals in LLM Safetycited by 0