Backdoor Defense in LLMs

LLM: Large Language Model

Ask about this topic

Overview

Backdoor defense studies how to find and neutralize hidden trigger-response behaviors planted through poisoned training data, fine-tuning, or model adapters. Methods include adversarial training and unlearning, weight or neuron repair, activation-based analysis, and runtime monitoring of signals such as entropy, attention, and hidden-state changes. Researchers evaluate these defenses on trigger-activated attack success as well as clean task performance, including in multimodal models, tool-using agents, and quantized or federated settings. A central challenge is defending against unknown or adaptive triggers without raising false alarms, damaging useful behavior, or adding unacceptable inference cost.

Latest papers

CardsList

42 papers12 papers in the last four weeks, up 300% on the four weeks before. 0.1% of all new papers.

  1. Interpretable sparse fine-tuning with TopKLoRA enables faithful circuit discovery

    Oct 23, 2026Marek Masiak, Andrzej Szablewski, Nicola Cancedda +1Sparse Fine-TuningLLM Interpretability

    Department of Computer Science, University of Cambridge · Neuralk-AI

  2. Poisoning-Based Backdoors Are Not Cryptographic

    Oct 20, 2026Masahiro KanekoBackdoor AttacksLLM Backdoor Attacks

    MBZUAI

  3. LLM Trojans that Increase Response Lengths and Cost

    Oct 17, 2026Zhengxing Li, Guangmingmei Yang, Nader Alfares +2LLM Backdoor AttacksAdversarial Attacks on LLMs

    Departments of Electrical Engineering and Computer Science & Engineering Penn State University University Park, PA 16802, USA

  4. Defending Against Dormant Poisoning Attacks Across Language and Multimodal Agents

    Oct 17, 2026Wonjun Lee, Vaidehi Patil, Zaid Khan +4AI Agent SafetyBackdoor Defense in LLMs

    UNC Chapel Hill · Kyung Hee University

  5. Beyond Static Cues: Unified Dynamic Detection of MLLM Backdoors via Dual-Flow Trajectory Analysis

    Oct 14, 2026Yinzhen Huang, Jiahe Xie, Yitong Sun +2Backdoor DetectionBackdoor Defense in LLMs

    Beihang University, Beijing, Beijing, China

  6. Forgetting is Not Always Bad: A Neuro-Inspired Memory Repair Mechanism for Poisoned LLM Agents

    Oct 9, 2026Lei Liu, Yunji Liang, Xiaowen Zhang +4Agent Memory PoisoningBackdoor Defense in LLMs

    Northwestern Polytechnical University · National Univerisity of Singapore

  7. Evaluating and Improving the Robustness of Large Language Models to Input Sequence Variations

    Oct 1, 2026Narek MaloyanLanguage Model RobustnessAdversarial Attacks on LLMs

  8. Backdoor Purification for LoRA-Tuned LLMs via Null-Space Projection

    Sep 30, 2026Jianwei Li, Jung-Eun KimLLM Backdoor AttacksBackdoor Defense in LLMs

    North Carolina State University Raleigh, NC 27606, USA

  9. Backdoor Containment via Expert Quarantine and Shutdown in LLMs

    Sep 30, 2026Jianwei Li, Min-Seon Kim, Jung-Eun KimLLM Backdoor AttacksBackdoor Defense in LLMs

    Department of Computer Science North Carolina State University Raleigh, NC 27606, USA

  10. Removing the NEEDLE in the Haystack: Backdoor Removal in LLMs via Weight Orthogonalisation

    Sep 29, 2026Minoo Kim, Vasileios Lampos, George DraysonBackdoor Defense in LLMsLLM Backdoor Attacks

    Locai Labs · Centre for AI, Computer Science, UCL

  11. Backdoor Mitigation in Decentralized LLM Fine-Tuning

    Sep 29, 2026Sayan Biswas, Jade Garcia Bourrée, Rachid Guerraoui +5Backdoor Defense in LLMsLLM Backdoor Attacks

    EPFL · University of California, San Diego

  12. Don't Inoculate Everything: Stratified Inoculation Prompting Narrows Backdoor Triggers and Preserves Desired Traits

    Sep 28, 2026Kajetan Dymkiewicz, Tim Farrelly, Adam Prada +7LLM Backdoor AttacksBackdoor Defense in LLMs

    University of Cambridge · Independent · University of California, Berkeley +3

  13. ServeGuard: Verifiable, Bounded-Residual Confinement of Operator-Invisible Channels Without Revealing the Certified Read Factor

    Sep 18, 2026Dominik Dahlem, Rui VieiraZero-Knowledge MLLLM Security

    Red Hat AI

  14. Universal Defenses for Tool-Integrated LLM Agents Against Adversarial Attacks

    Sep 14, 2026Xiaoyan Li, Yunli WangAdversarial Attacks on LLMsBackdoor Defense in LLMs

    Department of Computer Science University of Toronto Toronto, Ontario, Canada · Digital Technologies Research Centre National Research Council Canada Ottawa, Ontario, Canada

  15. SpecGuard: Inference-Time Backdoor Detection For Free

    Sep 10, 2026Rui Wen, Ahmed Salem, Andrew Paverd +2Backdoor DetectionBackdoor Defense in LLMs

    Institute of Science Tokyo · Microsoft Security Response Center · Microsoft Azure +1

  16. Not All Tokens Are Equal: Region-Aware Consistency Repair of Backdoors in MLLMs

    Aug 25, 2026Jiali Wei, Ming Fan, Mingkun Zhang +6Backdoor Attacks on VLMsMultimodal Large Language Models

    Xi’an Jiaotong University · Singapore Management University

  17. When Explanations Betray Backdoors: Black-Box Auditing for Language Model Classifiers

    Aug 12, 2026Yang Liu, Ran ZouLanguage Model AuditingLLM Backdoor Attacks

    Department of Statistics and Operations Research, University of North Carolina at Chapel Hill · Department of Statistics, University of California, Irvine

  18. Backdoor Decontamination Dynamics in LLM Agents

    Aug 11, 2026Gabriel Huang, Abhay Puri, Léo Boisvert +4LLM Backdoor AttacksBackdoor Defense in LLMs

    ServiceNow Research · Mila – Quebec AI Institute · Polytechnique Montréal +3

  19. LoRAScan: Detecting Backdoor Prompts in Low-Rank Adapters for Large Language Models via Down-Projection Activation Spikes

    Aug 7, 2026Doniyorkhon Obidov, Honggang Yu, Xiaolong Guo +1LLM Backdoor AttacksBackdoor Detection

    Michigan Technological University · Miami University · Lehigh University

  20. TriShield: Zero-Utility-Loss Defense Against Privacy Backdoors in Federated Language Model Fine-Tuning via Orthogonal Gradient Projection and Optimizer State Entanglement

    Jul 30, 2026Cheng WeiBackdoor AttacksBackdoor Attacks in Federated Learning

    Honor Device Co., Ltd., Shenzhen, China

  21. ToxScreen: Detecting Whether an LLM Has Been Poisoned

    Jul 29, 2026Anthony Hughes, Nicole Xing, Collin Francel +2LLM Backdoor AttacksBackdoor Detection

    University of Sheffield · Yale University · University of Alabama +2

  22. Defense Against LLM Backdoors using Critical Neuron Isolation Pruning

    Jul 22, 2026Yuxi Li, Zhibo Zhang, Kailong Wang +3LLM Backdoor AttacksBackdoor Defense in LLMs

    Huazhong University of Science and Technology China · Nanjing University of Aeronautics and Astronautics China · Nanyang Technological University Singapore

  23. Curvature-Guided Module Localization for Low-Rank Detoxification of Backdoored Large Language Models

    Jun 29, 2026Arash Raftari, Mehrdad Mahdavi, Nathan Blackthorn +1LLM Backdoor AttacksBackdoor Defense in LLMs

    AIVault Inc. Orlando, USA

  24. FlipGuard: Defending Large Language Models Against Quantization-Conditioned Backdoor Attacks

    Jun 27, 2026Aoying Zheng, Anqi Du, Zizhuang Deng +1LLM QuantizationLLM Backdoor Attacks

    School of Cyber Science and Technology Shandong University · Suzhou Research Institute of Shandong University

  25. Dummy Backdoor as a Defense: Removing Unknown Backdoors via Shared Internal Mechanisms for Generative LLMs

    Jun 10, 2026Kazuki Iwahana, Masaru Matsubayashi, Takuma Koyama +3LLM Backdoor AttacksBackdoor Defense in LLMs

    NTT Social Informatics Laboratories · Tohoku University

  26. Shared Latent Structures Enable Unified Backdoor Detection and Mitigation in LLMs

    Jun 6, 2026Omar Mahmoud, Aly M. Kassem, Thommen George Karimpanal +4LLM Backdoor AttacksBackdoor Defense in LLMs

    Applied Artificial Intelligence Initiative, Deakin University, Australia · Mila, Quebec AI Institute, Quebec, Canada · School of Information Technology, Deakin University, Australia +1

  27. Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs

    Jun 2, 2026Lisa Bouger, Théo Lasnier, Philippe Loubet Moundi +2LLM Backdoor AttacksBackdoor Defense in LLMs

    1Inria Paris · 2Sorbonne Université · 3Thales CDI

  28. Patcher: Post-Hoc Patching of Backdoored Large Language Models

    Jun 2, 2026Anjun Gao, Yueyang Quan, Yufei Xia +2LLM Backdoor AttacksBackdoor Attacks

    University of Louisville · University of North Texas

  29. Token-Level Generalization in LoRA Adapter Backdoors: Attack Characterization and Behavioral Detection

    May 28, 2026Travis LelleLLM Backdoor AttacksBackdoor Detection

  30. Be Kind, Rewrite: Benign Projections via Rewriting Defend Against LLM Data Poisoning Attacks

    May 18, 2026John T. Halloran, Noopur S. BhattLLM Backdoor AttacksTraining Data Curation

    Leidos · University of Pennsylvania