HARC: Coupling Harmfulness and Refusal Directions for Robust Safety Alignment
Understanding how aligned LLMs internally represent safety is critical for diagnosing alignment vulnerabilities, as it explains why jailbreaks succeed and informs the design of robust alignment strategies. Prior work shows that aligned LLMs encode harmfulness and refusal as separable directions in the residual stream at prompt-side token positions. We show that jailbreaks succeed at prompt encoding by suppressing either the refusal or harmfulness direction before any token is generated, with dis
Record details
Published: 1 July 2026
Source: arXiv
Category: Research
Topics: Safety & alignment · Healthcare
Retrieved: 14 July 2026
Related evidence
These records share source-supplied organisations, an exact publisher byline, automatic topics or regions. The reason is shown on every link; related does not mean supporting, agreeing with or verifying this record.
Multilayer Q-Matrix-Embedded Neural Network for Cognitive Diagnosis (M-QCDNet): Structure-Aware Deep Learning Architecture for Psychometric Interpretability
arXiv fairness query · 1 July 2026
Resolving superposition in AI for interpretability and cross-modal alignment in patient-neuronal images
arXiv · 30 June 2026
MolSight: A Graph-Aware Vision-Language Model for Unified Chemical Image Understanding
arXiv · 2 July 2026
RadiomicNet: A Hybrid Radiomics-Guided Lightweight Architecture for Interpretable Medical Image Segmentation
arXiv · 2 July 2026
When Aggregate Alignment Misleads: Auditing Policy Repair Without Per-State Expert Actions
arXiv · 3 July 2026
SonoCLIP: Mask-Guided Region-Aware Vision-Language Pretraining for Fetal Ultrasound Analysis
arXiv · 28 June 2026
How to cite this record
ethics.ai (1 July 2026), “HARC: Coupling Harmfulness and Refusal Directions for Robust Safety Alignment,” evidence record 362, https://ethics.ai/record/362 (originally published by arXiv).
Use and limitations
This page is a stable index and citation surface for a source record. ethics.ai did not author the underlying report and has not independently verified every claim. Automatic topics may be imperfect. For consequential use, quote and cite the original publisher.