English
Related papers

Related papers: Why Is RLHF Alignment Shallow? A Gradient Analysis

200 papers

We present a theoretical framework showing that popular LLM alignment methods, including RLHF and its variants, can be understood as divergence estimators between aligned (safe or preferred) and unaligned (harmful or less preferred)…

Machine Learning · Computer Science 2025-10-22 Rajdeep Haldar , Ziyi Wang , Qifan Song , Guang Lin , Yue Xing

Recent studies on the safety alignment of large language models (LLMs) have revealed that existing approaches often operate superficially, leaving models vulnerable to various adversarial attacks. Despite their significance, these studies…

Cryptography and Security · Computer Science 2025-06-02 Jianwei Li , Jung-Eun Kim

The safety alignment of current Large Language Models (LLMs) is vulnerable. Relatively simple attacks, or even benign fine-tuning, can jailbreak aligned models. We argue that many of these vulnerabilities are related to a shared underlying…

Cryptography and Security · Computer Science 2024-06-11 Xiangyu Qi , Ashwinee Panda , Kaifeng Lyu , Xiao Ma , Subhrajit Roy , Ahmad Beirami , Prateek Mittal , Peter Henderson

Large language models require continuous adaptation to new tasks while preserving safety alignment. However, fine-tuning on even benign data often compromises safety behaviors, including refusal of harmful requests, truthfulness, and…

Machine Learning · Computer Science 2026-04-21 Thong Bach , Dung Nguyen , Thao Minh Le , Truyen Tran

Safety alignment for large language models (LLMs) aims to reduce harmful or unsafe behavior while preserving general utility. However, recent findings reveal that alignment effects can be fragile: lightweight post-alignment manipulations,…

Artificial Intelligence · Computer Science 2026-05-29 Zhihao Liu , Yifan Wu , Jian Lou , Di Wang , Yuxi Zhou , Yuke Hu

Safety post-training can improve the harmfulness and policy compliance of Large Language Models (LLMs), but it may also reduce general utility, a phenomenon often described as the \emph{alignment tax}. We study this trade-off through the…

Machine Learning · Computer Science 2026-05-13 Guanglong Sun , Siyuan Zhang , Liyuan Wang , Jun Zhu , Hang Su , Yi Zhong

Large language models exhibit systematic vulnerabilities to adversarial attacks despite extensive safety alignment. We provide a mechanistic analysis revealing that position-dependent gradient weakening during autoregressive training…

Machine Learning · Computer Science 2025-11-18 Thong Bach , Dung Nguyen , Thao Minh Le , Truyen Tran

Large language models (LLMs) are now ubiquitous in everyday tools, raising urgent safety concerns about their tendency to generate harmful content. The dominant safety approach -- reinforcement learning from human feedback (RLHF) --…

Machine Learning · Computer Science 2025-09-29 Sathwik Karnik , Somil Bansal

Fine-tuning safety-aligned large language models (LLMs) can substantially compromise their safety. Previous approaches require many safety samples or calibration sets, which not only incur significant computational overhead during…

Machine Learning · Computer Science 2026-01-07 Jiawen Zhang , Lipeng He , Kejia Chen , Jian Lou , Jian Liu , Xiaohu Yang , Ruoxi Jia

Large language models (LLMs) have emerged as powerful tools for addressing a wide range of general inquiries and tasks. Despite this, fine-tuning aligned LLMs on smaller, domain-specific datasets, critical to adapting them to specialized…

Artificial Intelligence · Computer Science 2025-02-04 Guanlin Li , Kangjie Chen , Shangwei Guo , Jie Zhang , Han Qiu , Chao Zhang , Guoyin Wang , Tianwei Zhang , Jiwei Li

Fine-tuning large language models (LLMs) based on human preferences, commonly achieved through reinforcement learning from human feedback (RLHF), has been effective in improving their performance. However, maintaining LLM safety throughout…

Artificial Intelligence · Computer Science 2025-02-18 Yingshui Tan , Yilei Jiang , Yanshi Li , Jiaheng Liu , Xingyuan Bu , Wenbo Su , Xiangyu Yue , Xiaoyong Zhu , Bo Zheng

Reinforcement Learning from Human Feedback (RLHF) has become the predominant approach for language model (LM) alignment. At its core, RLHF uses a margin-based loss for preference optimization, specifying ideal LM behavior only by the…

Machine Learning · Computer Science 2025-04-23 Hui Yuan , Yifan Zeng , Yue Wu , Huazheng Wang , Mengdi Wang , Liu Leqi

Reinforcement learning (RL) has become a central post-training paradigm for large language models (LLMs), but its performance is highly sensitive to the quality of training problems. This sensitivity stems from the non-stationarity of RL:…

Machine Learning · Computer Science 2026-02-26 Ningyuan Yang , Weihua Du , Weiwei Sun , Sean Welleck , Yiming Yang

Large Language Models (LLMs) pose a significant risk of safety misalignment after finetuning, as models can be compromised by both explicitly and implicitly harmful data. Even some seemingly benign data can inadvertently steer a model…

Computation and Language · Computer Science 2026-05-15 Zhanhao Hu , Xiao Huang , Patrick Mendoza , Emad A. Alghamdi , Basel Alomair , Raluca Ada Popa , David Wagner

Large language models (LLMs) excel in various capabilities but pose safety risks such as generating harmful content and misinformation, even after safety alignment. In this paper, we explore the inner mechanisms of safety alignment through…

Computation and Language · Computer Science 2025-10-24 Jianhui Chen , Xiaozhi Wang , Zijun Yao , Yushi Bai , Lei Hou , Juanzi Li

Fine-tuning-as-a-Service introduces a critical vulnerability where a few malicious examples mixed into the user's fine-tuning dataset can compromise the safety alignment of Large Language Models (LLMs). While a recognized paradigm frames…

Computation and Language · Computer Science 2025-08-12 Biao Yi , Jiahao Li , Baolei Zhang , Lihai Nie , Tong Li , Tiansheng Huang , Zheli Liu

Fine-tuning aligned language models on benign tasks unpredictably degrades safety guardrails, even when training data contains no harmful content and developers have no adversarial intent. We show that the prevailing explanation, that…

Reinforcement Learning frameworks, particularly those utilizing human annotations, have become an increasingly popular method for preference fine-tuning, where the outputs of a language model are tuned to match a certain set of behavioral…

Machine Learning · Computer Science 2025-10-21 Archie Chaudhury

Alignment via reinforcement learning from human feedback (RLHF) has become the dominant paradigm for controlling the quality of outputs from large language models (LLMs). However, existing theories do not provide strong justification for…

Machine Learning · Computer Science 2026-05-19 Jihun Yun , Juno Kim , Jongho Park , Junhyuck Kim , Jongha Jon Ryu , Jaewoong Cho , Kwang-Sung Jun

Alignment in large language models (LLMs) is used to enforce guidelines such as safety. Yet, alignment fails in the face of jailbreak attacks that modify inputs to induce unsafe outputs. In this paper, we introduce and evaluate a new…

Cryptography and Security · Computer Science 2026-02-19 Jean-Charles Noirot Ferrand , Yohan Beugin , Eric Pauley , Ryan Sheatsley , Patrick McDaniel
‹ Prev 1 2 3 10 Next ›