LLM Safety Alignment

LLM: Large Language Model

Follow this topic Ask about this topic

Momentum

25 papers in the last four weeks, up 14% on the four weeks before. 0.2% of all new papers.

Jul 13Week of Sep 28

Latest papers 236

All topics

HARC: Coupling Harmfulness and Refusal Directions for Robust Safety Alignment
Jul 1, 2026Shei Pern Chua, Hao Wu, Qianli Ma +1Adversarial Attacks on LLMs LLM Safety Alignment
Tsinghua University · Microsoft · Southeast University
Revocable Learned State via Process Sidecars
Jun 29, 2026John SweeneyLLM Safety Alignment Task Arithmetic
Sideplane AI
EvalSafetyGap: A Hybrid Survey and Conceptual Framework for LLM Evaluation-Safety Failures
Jun 29, 2026Buğra Alperen Uluırmak, Rifat KurbanAI Alignment LLM Safety Alignment
Erciyes University, Türkiye. · Abdullah Gül University, Türkiye.
The Joint Effect of Quantization and Sampling Temperature on LLM Safety Alignment: A Factorial Analysis
Jun 28, 2026Hari Prasad, Ritam PalLLM Safety Benchmarks LLM Quantization
Conscious Engines
A Gravitational Interpretation of Safety Reversion under Fine-Tuning
Jun 26, 2026Samuele Poppi, Nils LukasFine-Tuning LLM Safety Alignment
Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), UAE
Robust Harmful Features Under Jailbreak Attacks: Mechanistic Evidence from Attention Head Specialization in Large Language Models
Jun 26, 2026Yanchen Yin, Dongqi Han, Linghui LiAttention Head Analysis Jailbreak Attacks
Beijing University of Posts and Telecommunications, Beijing, China.
Low-Agreeableness Persona Conditioning for Safe LLM Fine-Tuning
Jun 26, 2026Austin MY Cheung, Yi YangLLM Fine-Tuning LLM Safety Alignment
Hong Kong University of Science and Technology
TRACE: Trajectory-Based Safety Patch Learning for LLM Post-Training Realignment
Jun 26, 2026Changyue Li, Jiaming He, Youliang Yuan +4Supervised Fine-Tuning LLM Safety Alignment
The Chinese University of Hong Kong, Shenzhen · Ant Group · Lero the Research Ireland Centre for Software, University of Limerick
Yuvion LLM: An Adversarially-Aware Large Language Model for Content And AI Safety
Jun 26, 2026Ting Ma, Xiufeng Huang, Benlei Cui +43LLM Safety Benchmarks LLM Safety Alignment
RAS: Measuring LLM Safety Through Refusal Alignment
Jun 24, 2026Chang-Chieh Huang, Yan-Lun Chen, Chia-Mu Yu +1LLM Safety Alignment LLM Safety
National Yang Ming Chiao Tung University · Hon Hai Research Institute
PolicyAlign: Direct Policy-Based Safety Alignment for Large Language Models
Jun 24, 2026Chang Wu, Junfeng Fang, Houcheng Jiang +5LLM Alignment LLM Safety Alignment
Qwen Large Model Application Team, Alibaba · National University of Singapore · Zhongguancun Academy
Reinforcement Learning Towards Broadly and Persistently Beneficial Models
Jun 22, 2026Akshay V. Jagadeesh, Rahul K. Arora, Khaled Saab +5LLM Alignment LLM Safety Alignment
OpenAI
The Geometry of Refusal: Linear Instability in Safety-Aligned LLMs
Jun 21, 2026Shivam Ratnakar, Kartikeya VatsLanguage Model Steering Adversarial Attacks on LLMs
University of Southern California · Independent Researcher
Skin-Deep: A Geometric Diagnostic for Alignment Fragility in Large Language Model Representations
Jun 21, 2026Dongyub Jude Lee, Jungseob Lee, Seungyoon Lee +5LLM Interpretability LLM Fine-Tuning
Zoom Communications · Korea University · Yonsei University +1
Beyond Safe Data: Pretraining-Stage Alignment with Regular Safety Reflection
Jun 17, 2026Jinhan Li, Kexian Tang, Yihan Xu +2Language Model Pretraining LLM Alignment
Institute for Interdisciplinary Information Sciences, Tsinghua University
Self-CTRL: Self-Consistency Training with Reinforcement Learning
Jun 16, 2026Itamar Pres, Laura Ruis, Melat Ghebreselassie +2RL for Language Models Faithfulness of Language Model Explanations
MIT CSAIL
AnchorKV: Safety-Aware KV Cache Compression via Soft Penalty with a Refusal Anchor
Jun 16, 2026Ning Ni, Yingjie LaoKV Caching LLM Safety Alignment
Department of Computer Science Tufts University · Department of Electrical and Computer Engineering Tufts University
Adaptive and Explicit safe: Triggering Latent Safety Awareness in Large Reasoning Models
Jun 15, 2026Ke Miao, Jiaxin Li, Hongliang Chen +2LLM Safety Alignment LLM Safety
The State Key Laboratory of Blockchain and Data Security, Zhejiang University · Hangzhou HighTech Zone (Binjiang) Blockchain and Data Security Research Institute, China · Li Auto Inc. +2
SHARD: Safe and Helpful Alignment via Self-Reframing Distillation
Jun 14, 2026Viswonathan Manoranjan, Amogh Gupta, Anvesh Rao Vijjini +2LLM Safety Alignment Language Model Distillation
UNC Chapel Hill
Right or Wrong, Models Comply: Directional Blindness in LLM Moral Judgment
Jun 12, 2026Jihye Kim, Jeffrey FlaniganLLM Alignment Moral Reasoning in Language Models
University of California, Santa Cruz
A Virtuous AI is an Existential Risk
Jun 11, 2026Guillermo Del Pinal, Youngchan Lee, Min OhnLLM Safety Alignment AI Agent Safety
University of Massachusetts Amherst
ALIGNBEAM : Inference-Time Alignment Transfer via Cross-Vocabulary Logit Mixing
Jun 10, 2026Chirag Chawla, Pratinav Seth, Vinay Kumar SankarapuLLM Alignment LLM Safety Alignment
Lexsi Labs
Grammar-Constrained Decoding Can Jailbreak LLMs into Generating Malicious Code
Jun 10, 2026Yitong Zhang, Shiteng Lu, Jia LiLLM Safety Alignment LLM Jailbreak Attacks
College of AI Tsinghua University Beijing, China · Yingcai Honors College University of Electronic Science and Technology of China Chengdu, China
Training LLMs to Enforce Multi-Level Instruction Hierarchies via Gravity-Weighted Direct Preference Optimization
Jun 9, 2026Lena S. Bolliger, Lena A. JägerDirect Preference Optimization LLM Safety Alignment
Department of Computational Linguistics, University of Zurich, Switzerland
Emergent alignment and the projectability of ethical personas
Jun 8, 2026Guillermo Del Pinal, Youngchan Lee, Calum McNamara +1Moral Reasoning in Language Models Language Model Safety Evaluation
University of Massachusetts Amherst · Indiana University Bloomington
Distilling Safe LLM Systems via Soft Prompts for On Device Settings
Jun 8, 2026Motasem Alfarra, Cristina Pinneri, Dana Kianfar +2On-Device Language Model Inference LLM Safety Alignment
Qualcomm AI Research
Emergent Misalignment Can Be Induced by Sycophancy and Reversed via Alignment Gating
Jun 8, 2026Sicheng Wang, Xiangyang Zhu, Han Wang +6LLM Sycophancy LLM Alignment
Shanghai AI Lab, Shanghai, China.
LCAM: A Framework for Diagnosing Interactional Alignment Failures in Con-versational AI
Jun 6, 2026Manuele Reani, Hongyu TianLLM Alignment LLM Safety Alignment
School of Management and Economics, The Chinese University of Hong Kong, Shenzhen, China
Korean Culture into LLM Alignment: Toward Cultural Coherence
Jun 5, 2026MinJae Jung, Minwoo KimLLM Fine-Tuning LLM Safety Alignment
DATUMO Inc.
DOG-DPO:Dynamic Optimization in Geometry for Safety Alignment
Jun 4, 2026Yi Nian, Tiankai Yang, Yudi Zhang +7Direct Preference Optimization LLM Safety Alignment
University of Southern California · Iowa State University · University of Wisconsin–Madison +4

Jul 1, 2026, cs.AIJ/K move · Enter open · S save

HARC: Coupling Harmfulness and Refusal Directions for Robust Safety Alignment

Shei Pern Chua, Hao Wu, Qianli Ma, Fangzhao Wu

Tsinghua University · Microsoft · Southeast University

Adversarial Attacks on LLMs LLM Safety Alignment LLM Safety Adversarial Robustness of Language Models

Open paper AskPDF arXiv

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 distinct attack classes occupying separable regions of the harmfulness-refusal plane. Extending the analysis to response-token positions, we find that the model recognizes harmful content while it is generating that content, even when it failed to recognize the input as harmful at the prompt side. Motivated by our findings, we introduce HARC (Harmfulness-And-Refusal Coupling), a fine-tuning method that pairs the two directions across both prompt and response positions. Since the intervention is confined to the harmfulness-refusal subspace, it leaves the rest of the residual stream intact and does not degrade general capability or inflate over-refusal. Across extensive experiments, HARC achieves the strongest robustness-capability-usability trade-off among six baselines spanning the major training-time and inference-time safety methods. The harmfulness and refusal directions at prompt and response positions transfer across the five model families and two scales we tested without architecture-specific tuning.

Momentum

Related topics

Latest papers 236