Skip to main content
LLM Security Database
Skip to research search
Last analyzed Sep 9, 2026

Language Model Security Database

985 research findings · 1119 evaluated models

Filtered research findings

203 entries

Matches every word across titles, descriptions, sources, affected systems, and models.

LlamaGuard Obfuscation Bypass

Published Analyzed Source: arXiv

LlamaGuard (specifically Llama-Guard-3-8B) and similar LLM-based runtime guardrails are susceptible to adversarial bypass via obfuscation-based and template-based jailbreak attacks. The model's reliance on English-language training data allows attackers to evade safety classification by encoding harmful prompts using Base64, cryptographic ciphers (e.g., Caesar Cipher), or translating them into low-resource languages (e.g., Zulu). Furthermore, the model lacks sufficient alignment against…

DecipherGuard: Understanding and Deciphering Jailbreak Prompts for a Safer Deployment of Intelligent Software Systems
Evaluated models: Llama 3 8B

GUI Agent Dark Pattern Blindness

Published Analyzed Source: arXiv

Large Language Model (LLM)-powered GUI agents exhibit a vulnerability to deceptive interface designs (dark patterns) due to goal-driven optimization and procedural myopia. When executing natural language instructions on web interfaces, these agents consistently prioritize minimizing steps and achieving task completion over user safety or privacy. Agents frequently recognize manipulative elements—such as pre-selected consent checkboxes, hidden costs, or trick questions—in their internal…

Dark Patterns Meet GUI Agents: LLM Agent Susceptibility to Manipulative Interfaces and the Role of Human Oversight
Evaluated models: GPT-4o, Claude 3.7 Sonnet, DeepSeek V3+1 more
  • GPT-4o
  • Claude 3.7 Sonnet
  • DeepSeek V3
  • Gemini 2.0 Flash

Paper Submission Prompt Injection

Published Analyzed Source: arXiv

Large Language Models (LLMs) employed as automated assistants or autonomous agents in academic peer review systems are vulnerable to indirect prompt injection via maliciously crafted PDF submissions. Attackers can embed adversarial instructions within the manuscript that are invisible to human reviewers (using techniques such as white-on-white text or manipulating TrueType font character mapping tables) but are parsed and executed by the LLM.

When your reviewer is an llm: Biases, divergence, and prompt injection risks in peer review
Evaluated models: GPT-4o, GPT-5

Adversarial RAG Context Poisoning

Published Analyzed Source: arXiv

Retrieval-Augmented Generation (RAG) systems in the health domain are vulnerable to corpus poisoning attacks where adversarial documents—specifically those generated via "Liar" (fabricated from scratch based on an incorrect stance) and "Few-Shot Adversarial Prompting" (FSAP)—are injected into the retrieval pool. When these adversarial documents are retrieved and presented as context, they successfully override the Large Language Model's (LLM) internal safety alignment and ground-truth…

Evaluating the Robustness of Retrieval-Augmented Generation to Adversarial Evidence in the Health Domain
Evaluated models: GPT-4.1, GPT-5, Claude 3.5 Haiku+3 more
  • GPT-4.1
  • GPT-5
  • Claude 3.5 Haiku
  • DeepSeek R1 Distill Qwen 32B
  • Phi-4
  • Llama 3 8B Instruct

Knowledge Neuron Jailbreak

Published Analyzed Source: arXiv

Aligned Large Language Models (LLMs) utilizing Transformer architectures are vulnerable to representation-level attacks targeting safety-knowledge neurons within the Multi-Layer Perceptron (MLP) layers. Research indicates that safety decision-making (Rejection vs. Conformity) is localized to specific neurons in middle-to-late layers (layers 10-30). An attacker with white-box access can calculate a "Conformity" direction vector based on the activation differences between benign and harmful…

Unraveling LLM Jailbreaks Through Safety Knowledge Neurons
Evaluated models: Llama 2 7B, Vicuna 7B

Malicious Intent Bypass

Published Analyzed Source: arXiv

IntentionReasoner, specifically the 1.5B and 3B parameter versions optimized via Reinforcement Learning (RL), contains a safety regression vulnerability where the RL alignment process degrades the model's resistance to jailbreak attacks compared to the Supervised Fine-Tuning (SFT) baseline. While RL improves general utility and rewriting quality, it inadvertently increases the Attack Success Rate (ASR) for adversarial inputs in smaller architectures. This allows sophisticated jailbreak prompts…

IntentionReasoner: Facilitating Adaptive LLM Safeguards through Intent Reasoning and Selective Query Refinement
Evaluated models: GPT-4o, Qwen 2.5 7B Instruct, Llama 3.1 8B Instruct+3 more
  • GPT-4o
  • Qwen 2.5 7B Instruct
  • Llama 3.1 8B Instruct
  • DeepSeek V3
  • IntentionReasoner 1.5B
  • IntentionReasoner 3B

Multimodal Linking Visual Insecurity

Published Analyzed Source: arXiv

Multimodal Entity Linking (MEL) systems, encompassing both traditional dual-encoder models and Multimodal Large Language Models (MLLMs), are vulnerable to gradient-based white-box adversarial attacks. By applying imperceptible perturbations to visual inputs via Projected Gradient Descent (PGD), Auto-PGD (APGD), or Carlini & Wagner (CW) methods, an attacker can manipulate the visual embeddings generated by the model. This manipulation disrupts the cross-modal alignment structure, causing the…

On Evaluating the Adversarial Robustness of Foundation Models for Multimodal Entity Linking
Evaluated models: MiniGPT-4

Back-Translation Watermark Stripping

Published Analyzed Source: arXiv

Implementations of Large Language Model (LLM) watermarking algorithms—specifically KGW (Kirchenbauer et al.), Semantic Invariant Robust (SIR) Watermark, Entropy-based Text Watermarking (EWD), and Unbiased Watermarking—are vulnerable to watermark stripping via adversarial text perturbation. When watermarked text generated by models such as OPT-1.3B is subjected to automated paraphrasing or back-translation (e.g., English $\to$ French $\to$ English), the embedded statistical signals are…

Signature vs. Substance: Evaluating the Balance of Adversarial Resistance and Linguistic Quality in Watermarking Large Language Models
Evaluated models: Llama 3 8B

Physical Patch Driving Hijack

Published Analyzed Source: arXiv

Multimodal Large Language Models (MLLMs) employed in autonomous driving (AD) systems are vulnerable to a physically realizable adversarial patch attack dubbed "PhysPatch." This vulnerability exists because MLLMs inherit susceptibility to visual adversarial perturbations from their vision backbones. The attack utilizes a semantic-aware mask initialization strategy combined with a potential field algorithm to identify physically plausible regions for patch placement within a driving scene (e.g…

PhysPatch: A Physically Realizable and Transferable Adversarial Patch Attack for Multimodal Large Language Models-based Autonomous Driving Systems
Evaluated models: LLaVA v1.6 13B, Qwen 2.5 VL 72B Instruct, Llama 3.2 90B Vision Instruct+8 more
  • LLaVA v1.6 13B
  • Qwen 2.5 VL 72B Instruct
  • Llama 3.2 90B Vision Instruct
  • GPT-4o
  • GPT-4.1
  • Claude Sonnet 4
  • Gemini 2.0 Flash
  • Qwen 2.5 VL Max
  • o3
  • Gemini 2.5 Flash
  • QVQ-Plus

Graph-LLM Template Injection

Published Analyzed Source: arXiv

A vulnerability exists in the graph encoding architecture of LLaGA (Large Language and Graph Assistant), specifically within the "neighborhood detail template" used to construct node sequences. LLaGA enforces a fixed-shape computational tree for each node; when a target node has fewer neighbors than the required template size (e.g., $k$ children), the system utilizes placeholders to maintain the fixed structure.

Adversarial Attacks and Defenses on Graph-aware Large Language Models (LLMs)
Evaluated models: GPT-4, Llama 2 7B, Vicuna 7B

Research methodology

Entries summarize publicly available primary-source security research. Model names reflect only systems explicitly evaluated by the cited paper, and measurements are research-reported unless independent verification is stated.

Discovery and publication are separate: new papers enter a candidate queue, then require human review and explicit approval before publication. Updates follow that review process, not a guaranteed daily publishing schedule. “Last analyzed” is the newest analysis date among published entries; it does not mean that all recent research has been covered.