mlsanityData to 14 Oct 2026

Universal Refusal Circuits Across LLMs: Cross-Model Transfer via Trajectory Replay and Concept-Basis Reconstruction

arXiv 2601.16034 ↗First posted 22 Jan 2026cs.CLcited by 2

On the map: Mechanistic interpretability → Steering and probing for LLM safety

Abstract (arXiv)

Refusal behavior in aligned LLMs is often viewed as model-specific, yet we hypothesize it stems from a universal, low-dimensional semantic circuit shared across models. To test this, we introduce Trajectory Replay via Concept-Basis Reconstruction, a framework that transfers refusal interventions from donor to target models, spanning diverse architectures (e.g., Dense to MoE) and training regimes, without using target-side refusal supervision. By aligning layers via concept fingerprints and reconstructing refusal directions using a shared ``recipe'' of concept atoms, we map the donor's ablation trajectory into the target's semantic space. To preserve capabilities, we introduce a weight-SVD stability guard that projects interventions away from high-variance weight subspaces to prevent collateral damage. Our evaluation across 8 model pairs confirms that these transferred recipes consistently attenuate refusal while maintaining performance, providing strong evidence for the semantic universality of safety alignment.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Steering and probing for LLM safety, one of 13 directions in Mechanistic interpretability. Builds on 15 of its 42 references that are in the map. First posted on arXiv 22 Jan 2026.

Builds on

15 of 42 references.

References · 15

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  3. 2025Qwen3 Technical Reportcited by 8,820
  4. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  5. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  6. 2017Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layercited by 3,509
  7. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  8. 2016Pointer Sentinel Mixture Modelscited by 3,096
  9. 2021Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsitycited by 3,073
  10. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  11. 2019Similarity of Neural Network Representations Revisitedcited by 1,755
  12. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
3 more
  1. 2022Mass-Editing Memory in a Transformercited by 1,001
  2. 2024AlphaEdit: Null-Space Constrained Knowledge Editing for Language Modelscited by 236
  3. 2026Surgical Refusal Ablation: Disentangling Safety from Intelligence via Concept-Guided Spectral Cleaningcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Contrastive-Difference CKA Reveals Concept-Specific Structural Alignment Across Language Model Architecturescited by 0
  2. 2026Learning Cross-Model Activation Alignments with Explicit Many-to-Many Layer Mapscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026The Geometry of Refusal: Linear Instability in Safety-Aligned LLMs
  2. 2026Over-Refusal and Representation Subspaces: A Mechanistic Analysis of Task-Conditioned Refusal in Aligned LLMs
  3. 2026Efficient Refusal Ablation in LLM through Optimal Transport
  4. 2026Beyond Shallow Alignment: How Post-Training Methods Determine Refusal Circuits And Steering Robustness
  5. 2025SafeConstellations: Mitigating Over-Refusals in LLMs Through Task-Aware Representation Steering
  6. 2026LoMC: Localized Multidirectional Correction for Refusal Suppression in Routed Foundation Models
  7. 2026From Refusal Tokens to Refusal Control: Discovering and Steering Category-Specific Refusal Directions
  8. 2026Mitigating LLM Over-Refusal via Dynamic Semantic Routing Calibration