EigenShield: Inference-Time, Model-Agnostic Jailbreaking Defense via Causal Subspace Filtering
Nastaran Darabi, Devashri Naik, Sina Tayebati, Dinithi Jayasuriya, Ranganath Krishnan, Amit Ranjan Trivedi
Abstract
Large Language Models (LLMs) and Vision-Language Models (VLMs) remain highly vulnerable to adversarial attacks despite widespread adoption. Existing defenses typically require retraining, rely on heuristics, or fail under adaptive and out-of-distribution (OOD) conditions. We introduce EigenShield, a principled, inference-time, architecture-agnostic defense that leverages Random Matrix Theory (RMT) to suppress adversarial noise in high-dimensional embeddings. EigenShield uses spiked covariance modeling and a Robustness-based Nonconformity Score (RbNS) with quantile thresholding to isolate and preserve causal eigenvectors, filtering out adversarial components without model access or adversarial training. We develop a theoretical framework establishing conditions for asymptotic noise suppression and demonstrate effectiveness in both unimodal and multimodal settings. Empirically, EigenShield consistently improves robustness across threat models, reducing attack success rates (ASR) by up to 48% over state-of-the-art defenses, including adversarial training, UNIGUARD, CIDER, and input transformations. On jailbreak attacks, EigenShield lowers LLM ASR by up to 92.9% relative to undefended models. Under multimodal adversarial attacks, it reduces VLM ASR by up to 76.5%. Against adaptive attacks on LLMs, it achieves ASR reductions of up to 77.7%. In OOD settings, EigenShield maintains strong performance, reducing ASR by up to 88.4% for LLMs and 80.4% for VLMs.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning cites MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models
- HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal cites Towards Evaluating the Robustness of Neural Networks
Builds on11
- Visual Instruction TuningHaotian Liu, Chunyuan Li, Qingyang Wu, Yong Jae LeeNeurIPS 2023 · 11,349 citations
- Towards Evaluating the Robustness of Neural NetworksNicholas Carlini, David A. WagnerS&P 2017 · 9,786 citations
- InstructBLIP: Towards General-purpose Vision-Language Models with Instruction TuningWenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong et al.NeurIPS 2023 · 4,013 citations
- MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language ModelsDeyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li et al.ICLR 2024 · 3,079 citations
- HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust RefusalMantas Mazeika, Long Phan, Xuwang Yin, Andy Zou et al.ICML 2024 · 1,031 citations
Related papers
- MirrorShield: Towards Dynamic Adaptive Defense Against Jailbreaks via Entropy-Guided Mirror CraftingRui Pu, Chaozhuo Li, Rui Ha, Litian Zhang et al.AAAI 2026
- Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against JailbreaksHan Wang, Gang Wang, Huan ZhangCVPR 2025
- MetaDefense: Defending Fine-tuning based Jailbreak Attack Before and During GenerationWeisen Jiang, Sinno Jialin PanNeurIPS 2025 · 10 citations
- Attack as Defense: Safeguarding Large Vision-Language Models from Jailbreaking by Adversarial AttacksChongxin Li, Hanzhang Wang, Yuchun FangEMNLP Findings 2025 · 1 citation
- Understanding and Improving Continuous LLM Adversarial Training via In-context Learning TheoryShaopeng Fu, Di WangICLR 2026 · 1 citation
