English
Related papers

Related papers: Language-Switching Triggers Take a Latent Detour T…

200 papers

In the field of natural language processing, the prevalent approach involves fine-tuning pretrained language models (PLMs) using local samples. Recent research has exposed the susceptibility of PLMs to backdoor attacks, wherein the…

Machine Learning · Computer Science 2023-10-31 Ruixiang Tang , Jiayi Yuan , Yiming Li , Zirui Liu , Rui Chen , Xia Hu

Backdoor vulnerabilities widely exist in the fine-tuning of large language models(LLMs). Most backdoor poisoning methods operate mainly at the token level and lack deeper semantic manipulation, which limits stealthiness. In addition, Prior…

Computation and Language · Computer Science 2026-05-13 Ziyu Liu , Tao Li , Tianjie Ni , Xiaolong Lan , Wengang Ma , Tao Yang , Guohua Wang , Junjiang He

Recent research highlights concerns about the trustworthiness of third-party Pre-Trained Language Models (PTLMs) due to potential backdoor attacks. These backdoored PTLMs, however, are effective only for specific pre-defined downstream…

Cryptography and Security · Computer Science 2025-05-20 Rui Zhang , Yun Shen , Hongwei Li , Wenbo Jiang , Hanxiao Chen , Yuan Zhang , Guowen Xu , Yang Zhang

Large language models (LLMs) have seen significant advancements, achieving superior performance in various Natural Language Processing (NLP) tasks, from understanding to reasoning. However, they remain vulnerable to backdoor attacks, where…

Computation and Language · Computer Science 2024-11-28 Chen Chen , Yuchen Sun , Xueluan Gong , Jiaxin Gao , Kwok-Yan Lam

Textual backdoor attacks present a substantial security risk to Large Language Models (LLM). It embeds carefully chosen triggers into a victim model at the training stage, and makes the model erroneously predict inputs containing the same…

Computation and Language · Computer Science 2024-07-08 Xinglin Li , Xianwen He , Yao Li , Minhao Cheng

Recent studies have suggested a processing framework for multilingual inputs in decoder-based LLMs: early layers convert inputs into English-centric and language-agnostic representations; middle layers perform reasoning within an…

Computation and Language · Computer Science 2025-09-23 Hinata Tezuka , Naoya Inoue

Detecting whether a model has been poisoned is a longstanding problem in AI security. In this work, we present a practical scanner for identifying sleeper agent-style backdoors in causal language models. Our approach relies on two key…

Cryptography and Security · Computer Science 2026-02-04 Blake Bullwinkel , Giorgio Severi , Keegan Hines , Amanda Minnich , Ram Shankar Siva Kumar , Yonatan Zunger

Large reasoning models (LRMs) achieve strong performance on mathematical reasoning tasks, often attributed to their capability to generate explicit chain-of-thought (CoT) explanations. However, recent work shows that LRMs often arrive at…

Computation and Language · Computer Science 2026-04-20 Yihong Liu , Raoyuan Zhao , Hinrich Schütze , Michael A. Hedderich

Backdoor attacks are a kind of insidious security threat against machine learning models. After being injected with a backdoor in training, the victim model will produce adversary-specified outputs on the inputs embedded with predesigned…

Computation and Language · Computer Science 2021-06-04 Fanchao Qi , Mukai Li , Yangyi Chen , Zhengyan Zhang , Zhiyuan Liu , Yasheng Wang , Maosong Sun

Large language model (LLM) agents execute tasks through multi-step workflows that combine planning, memory, and tool use. While this design enables autonomy, it also expands the attack surface for backdoor threats. Backdoor triggers…

Artificial Intelligence · Computer Science 2026-01-13 Yunhao Feng , Yige Li , Yutao Wu , Yingshui Tan , Yanming Guo , Yifan Ding , Kun Zhai , Xingjun Ma , Yu-Gang Jiang

Backdoors are hidden behaviors that are only triggered once an AI system has been deployed. Bad actors looking to create successful backdoors must design them to avoid activation during training and evaluation. Since data used in these…

Cryptography and Security · Computer Science 2024-12-25 Sara Price , Arjun Panickssery , Sam Bowman , Asa Cooper Stickland

Supervised fine-tuning has become the predominant method for adapting large pretrained models to downstream tasks. However, recent studies have revealed that these models are vulnerable to backdoor attacks, where even a small number of…

Cryptography and Security · Computer Science 2025-01-08 Peihai Jiang , Xixiang Lyu , Yige Li , Jing Ma

Large language models (LLMs) often benefit from intermediate steps of reasoning to generate answers to complex problems. When these intermediate steps of reasoning are used to monitor the activity of the model, it is essential that this…

Machine Learning · Computer Science 2023-11-02 Fabien Roger , Ryan Greenblatt

Text generation has a fundamental limitation almost by definition: there is no taking back tokens that have been generated, even when they are clearly problematic. In the context of language model safety, when a partial unsafe generation is…

Machine Learning · Computer Science 2024-09-24 Yiming Zhang , Jianfeng Chi , Hailey Nguyen , Kartikeya Upasani , Daniel M. Bikel , Jason Weston , Eric Michael Smith

Multimodal Large Language Models (MLLMs) have achieved remarkable success in cross-modal understanding and generation, yet their deployment is threatened by critical safety vulnerabilities. While prior works have demonstrated the…

Cryptography and Security · Computer Science 2026-04-22 Kun Wang , Cheng Qian , Miao Yu , Lilan Peng , Liang Lin , Jiaming Zhang , Tianyu Zhang , Yu Cheng , Yang Wang

Trojan backdoors can be injected into large language models at various stages, including pretraining, fine-tuning, and in-context learning, posing a significant threat to the model's alignment. Due to the nature of causal language modeling,…

Computation and Language · Computer Science 2025-01-22 Vedant Bhasin , Matthew Yudin , Razvan Stefanescu , Rauf Izmailov

Modern language models remain vulnerable to backdoor attacks via poisoned data, where training inputs containing a trigger are paired with a target output, causing the model to reproduce that behavior whenever the trigger appears at…

Cryptography and Security · Computer Science 2026-01-06 Eric Xue , Ruiyi Zhang , Pengtao Xie

The use of third-party datasets and pre-trained machine learning models poses a threat to NLP systems due to possibility of hidden backdoor attacks. Existing attacks involve poisoning the data samples such as insertion of tokens or sentence…

Computation and Language · Computer Science 2024-04-09 Irina Alekseevskaia , Konstantin Arkhipenko

Multimodal pretrained models are vulnerable to backdoor attacks, yet most existing methods rely on visual or multimodal triggers, which are impractical since visually embedded triggers rarely occur in real-world data. To overcome this…

Cryptography and Security · Computer Science 2026-04-08 Yiyang Zhang , Chaojian Yu , Ziming Hong , Yuanjie Shao , Qinmu Peng , Tongliang Liu , Xinge You

Early research into data poisoning attacks against Large Language Models (LLMs) demonstrated the ease with which backdoors could be injected. More recent LLMs add step-by-step reasoning, expanding the attack surface to include the…

Cryptography and Security · Computer Science 2025-09-09 Hanna Foerster , Ilia Shumailov , Yiren Zhao , Harsh Chaudhari , Jamie Hayes , Robert Mullins , Yarin Gal