Word Level Attack

Word-level attacks aim to subtly alter text inputs to mislead natural language processing (NLP) models, revealing vulnerabilities in their robustness and raising concerns about their reliability in real-world applications. Current research focuses on developing both more effective attacks, often employing techniques like Markov decision processes and beam search to optimize perturbations, and stronger defenses, such as stochastic purification and adversarial training methods that learn more robust representations. Understanding and mitigating these vulnerabilities is crucial for ensuring the safety and trustworthiness of NLP models across various domains, from sentiment analysis to hate speech detection.

Papers

November 1, 2024

Plentiful Jailbreaks with String Compositions
Brian R.Y. Huang
Jailbreak Attack Obfuscation Technique Word Level Attack Optimal Attack Invertible Linear Transformation World War II Era Cipher

August 11, 2024

Kov: Transferable and Naturalistic Black-Box LLM Attacks using Markov Decision Processes and Tree Search
Robert J. Moss
Adversarial Attack Markov Decision Process Tree Search Black Box LLM Word Level Attack LLM Attack Token Level Attack

June 18, 2024

June 12, 2024

Adversarial Evasion Attack Efficiency against Large Language Models
João Vitorino, Eva Maia, Isabel Praça
Adversarial Attack Adversarial Example Adversarial Defense Word Level Attack Adversarial Evasion Attack

May 14, 2024

Stylometric Watermarks for Large Language Models
Georg Niess, Roman Kern
Large Language Model Shot Classification Linguistic Feature Word Level Attack Stylometric Feature

April 9, 2024

Sandwich attack: Multi-language Mixture Adaptive Attack on LLMs
Bibek Upadhayay, Vahid Behzadan
Large Language Model Attack Strategy Word Level Attack Sandwich Attack

March 27, 2024

SemRoDe: Macro Adversarial Training to Learn Representations That are Robust to Word-Level Attacks
Brian Formento, Wenjie Feng, Chuan Sheng Foo, Luu Anh Tuan, See-Kiong Ng
Adversarial Attack Adversarial Training Meaningful Representation Domain Adversarial Word Level Attack

November 29, 2023

SenTest: Evaluating Robustness of Sentence Encoders
Tanmay Chavan, Shantanu Patankar, Aditya Kane, Omkar Gokhale, Geetanjali Kale, Raviraj Joshi
Adversarial Attack Native Robustness Sentence Representation Sentence Encoders Word Level Attack Sentence Level Attack

October 4, 2023

Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models
Xianjun Yang, Xiao Wang, Qi Zhang, Linda Petzold, William Yang Wang, Xun Zhao, Dahua Lin
Large Language Model AI Safety Prompt Attack Shadow Generation Word Level Attack Aligned Model Alignment Breaking Attack

September 25, 2023

The Cybersecurity Crisis of Artificial Intelligence: Unrestrained Adoption and Natural Language-Based Attacks
Andreas Tsamados, Luciano Floridi, Mariarosaria Taddeo
Artificial Intelligence Natural Language Security Vulnerability Autoregressive Large Language Model Visual Naturalness Common Vulnerability Adoption Strategy Word Level Attack Cybersecurity Incident

September 4, 2023

MathAttack: Attacking Large Language Models Towards Math Solving Ability
Zihao Zhou, Qiufeng Wang, Mingyu Jin, Jie Yao, Jianan Ye, Wei Liu, Wei Wang, Xiaowei Huang, Kaizhu Huang
Adversarial Attack Critique Ability Level Mathematics Word Level Attack Robust Dataset

April 16, 2023

Language Guided Local Infiltration for Interactive Image Retrieval
Fuxiang Huang, Lei Zhang
Image Retrieval Visual Prompt Textual Feature Word Level Attack

March 28, 2023

TransAudio: Towards the Transferable Adversarial Audio Attack via Learning Contextualized Perturbations
Qi Gege, Yuefeng Chen, Xiaofeng Mao, Yao Zhu, Binyuan Hui, Xiaodan Li, Rong Zhang, Hui Xue
Adversarial Example Transferable Attack Adversarial Audio Word Level Attack Transfer Based Attack Context Perturbation

March 9, 2023

BeamAttack: Generating High-quality Textual Adversarial Examples through Beam Search and Mixed Semantic Spaces
Hai Zhu, Qingyang Zhao, Yuren Wu
Adversarial Example Adversarial Training Beam Search Semantic Space Textual Adversarial Example Word Level Attack Textual Attack

February 6, 2023

Less is More: Understanding Word-level Textual Adversarial Attack via n-gram Frequency Descend
Ning Lu, Shengcai Liu, Zhirui Zhang, Qi Wang, Haifeng Liu, Ke Tang
Adversarial Example Adversarial Training N Gram Textual Adversarial Attack Word Level Attack

May 24, 2022

Certified Robustness Against Natural Language Attacks by Causal Intervention
Haiteng Zhao, Chang Ma, Xinshuai Dong, Anh Tuan Luu, Zhi-Hong Deng, Hanwang Zhang
Native Robustness Adversarial Example Adversarial Vulnerability Causal Intervention Word Level Attack Synonym Substitution Attack

April 10, 2022

"That Is a Suspicious Reaction!": Interpreting Logits Variation to Detect NLP Adversarial Attacks
Edoardo Mosca, Shreyash Agarwal, Javier Rando, Georg Groh
Natural Language Processing Adversarial Attack Adversarial Input Textual Adversarial Example Word Level Attack Suspicious Commenter Adversarial Natural Language

February 11, 2022

White-Box Attacks on Hate-speech BERT Classifiers in German with Explicit and Implicit Character Level Defense
Shahrukh Khan, Mahnoor Shahid, Navdeeppal Singh
Adversarial Robustness BERT Model Hate Speech White Box Explicit Content Word Level Attack

Word Level Attack

Papers

Plentiful Jailbreaks with String Compositions

Kov: Transferable and Naturalistic Black-Box LLM Attacks using Markov Decision Processes and Tree Search

MaskPure: Improving Defense Against Text Adversaries with Stochastic Purification

Cross-Lingual Unlearning of Selective Knowledge in Multilingual Language Models

Adversarial Evasion Attack Efficiency against Large Language Models

Stylometric Watermarks for Large Language Models

Sandwich attack: Multi-language Mixture Adaptive Attack on LLMs

SemRoDe: Macro Adversarial Training to Learn Representations That are Robust to Word-Level Attacks

SenTest: Evaluating Robustness of Sentence Encoders

Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models

The Cybersecurity Crisis of Artificial Intelligence: Unrestrained Adoption and Natural Language-Based Attacks

MathAttack: Attacking Large Language Models Towards Math Solving Ability

Language Guided Local Infiltration for Interactive Image Retrieval

TransAudio: Towards the Transferable Adversarial Audio Attack via Learning Contextualized Perturbations

BeamAttack: Generating High-quality Textual Adversarial Examples through Beam Search and Mixed Semantic Spaces

Less is More: Understanding Word-level Textual Adversarial Attack via n-gram Frequency Descend

Certified Robustness Against Natural Language Attacks by Causal Intervention

"That Is a Suspicious Reaction!": Interpreting Logits Variation to Detect NLP Adversarial Attacks

White-Box Attacks on Hate-speech BERT Classifiers in German with Explicit and Implicit Character Level Defense