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.

Chinese Pattern Safety Evasion

Published Analyzed Source: arXiv

Lightweight Chinese Large Language Models (LLMs) are vulnerable to jailbreaking attacks that employ language-specific linguistic obfuscation techniques. Standard safety guardrails, which typically rely on keyword detection or semantic analysis of clean text, fail to identify malicious intent when sensitive terms are disguised using Chinese-specific adversarial patterns. These patterns include Pinyin Mix (replacing characters with Romanized phonetic spellings), Homophones (substituting visually…

CSSBench: Evaluating the Safety of Lightweight LLMs against Chinese-Specific Adversarial Patterns
Evaluated models: Qwen 3 0.6B, Qwen 3 1.7B, Qwen 3 8B+7 more
  • Qwen 3 0.6B
  • Qwen 3 1.7B
  • Qwen 3 8B
  • MiniCPM4 0.5B
  • MiniCPM4 8B
  • Hunyuan 0.5B
  • Hunyuan 1.8B
  • Hunyuan 7B
  • openPangu-Embedded 1B
  • openPangu-Embedded 7B

Single Word Video Corruption

Published Analyzed Source: arXiv

Text-to-Video (T2V) diffusion models are vulnerable to black-box adversarial prompt attacks that degrade output quality regarding semantic fidelity and temporal dynamics. This vulnerability is exploited via the T2VAttack framework, which utilizes two primary vector strategies: T2VAttack-S (Substitution) and T2VAttack-I (Insertion). T2VAttack-S leverages a greedy search to identify key semantic tokens and replaces them with high-similarity synonyms defined in lexical databases (e.g., WordNet)…

T2VAttack: Adversarial Attack on Text-to-Video Diffusion Models
Evaluated models: Not reported

Adversarial Vision Degrades VLM

Published Analyzed Source: arXiv

Large Language and Vision Assistant (LLaVA) v1.5-13B and Meta Llama 3.2 11B Vision are vulnerable to adversarial evasion attacks targeting the visual input modality. An attacker with white-box access (knowledge of model architecture and gradients) can employ Projected Gradient Descent (PGD) to generate adversarial perturbations constrained by an L-infinity norm. By maximizing the model's internal loss function with respect to the input image, the attacker can force the Vision-Language Model…

Adversarial Robustness of Vision in Open Foundation Models
Evaluated models: LLaVA 1.5 13B, Llama 3.2 11B Vision

LLM Infinite Thinking DoS

Published Analyzed Source: arXiv

A Denial-of-Service (DoS) vulnerability exists in Large Language Model (LLM) inference services where specially crafted input prompts can trigger excessively long or infinite generation loops ("infinite thinking"). This vulnerability, identified as "ThinkTrap," utilizes derivative-free optimization (CMA-ES) within a continuous surrogate embedding space to circumvent the discrete nature of token inputs. By optimizing a low-dimensional latent vector and projecting it to token sequences, an…

ThinkTrap: Denial-of-Service Attacks against Black-box LLM Services via Infinite Thinking
Evaluated models: Gemini 2.5 Pro, Lumimaid 70B, o4-mini+4 more
  • Gemini 2.5 Pro
  • Lumimaid 70B
  • o4-mini
  • MAI DS R1 671B
  • DeepSeek R1 0528 Qwen3 8B
  • Llama 3.2 3B
  • DeepSeek R1 671B

Guardrail Helpful Mode Jailbreak

Published Analyzed Source: arXiv

A "Helpful Mode" role-confusion vulnerability exists in specific Large Language Model (LLM) safety guardrails, specifically Nemotron-Safety-8B and Granite-Guardian-3.2-5B. These models, designed to act as binary classifiers (outputting "Safe" or "Unsafe") for content moderation, can be manipulated via contextually framed adversarial prompts (e.g., academic research requests, corporate security scenarios, or roleplay) to abandon their classification objective. Instead of blocking the request…

Evaluating the Robustness of Large Language Model Safety Guardrails Against Adversarial Attacks
Evaluated models: Nemotron Safety 8B, Granite Guardian 3.2 5B

Embodied Cross-Modal Misalignment

Published Analyzed Source: arXiv

OpenVLA, a Vision-Language-Action (VLA) model, contains a vulnerability regarding multimodal adversarial robustness. The model lacks sufficient cross-modal alignment stability, allowing attackers to disrupt the grounding between visual perception and linguistic instructions. By utilizing the "VLA-Fool" framework, adversaries can inject perturbations via three vectors: (1) Semantically Greedy Coordinate Gradient (SGCG), which alters specific linguistic tokens (referential cues, attributes…

When alignment fails: Multimodal adversarial attacks on vision-language-action models
Evaluated models: Not reported

Indirect Environmental Jailbreak

Published Analyzed Source: arXiv

Embodied Artificial Intelligence (AI) agents utilizing Vision-Language Models (VLMs) for perception and planning are vulnerable to Indirect Environmental Jailbreak (IEJ). The vulnerability arises from the system's failure to distinguish between user-issued instructions and text embedded in the physical environment (e.g., writing on walls, sticky notes, or projections). The VLM processes visual text detected in the camera feed as authoritative context or direct commands, allowing a black-box…

The Shawshank Redemption of Embodied AI: Understanding and Benchmarking Indirect Environmental Jailbreaks
Evaluated models: GPT-4o, Qwen3-VL Plus, Gemini 2.0 Flash+3 more
  • GPT-4o
  • Qwen3-VL Plus
  • Gemini 2.0 Flash
  • GLM 4.5
  • DeepSeek-VL2
  • Claude 3.5 Sonnet

Black-Box Graph-Text Node Injection

Published Analyzed Source: arXiv

LLM-enhanced Graph Neural Networks (GNNs), which integrate Large Language Model (LLM) feature encoders with graph message-passing architectures, are vulnerable to a black-box node injection attack known as "GraphTextack." This vulnerability exists because the joint model architecture creates a dual attack surface: the GNN component is sensitive to structural perturbations (changes in graph topology), while the LLM component is sensitive to semantic perturbations (adversarial phrasing).

GRAPHTEXTACK: A Realistic Black-Box Node Injection Attack on LLM-Enhanced GNNs
Evaluated models: Llama 2 7B

Multi-Agent Typo Vulnerability

Published Analyzed Source: arXiv

Multi-agent Large Language Model (LLM) systems employing ensemble sampling-and-voting strategies (specifically the "Agent Forest" framework) are vulnerable to adversarial input perturbations. While increasing the number of agents ($n \in \{1, \dots, 25\}$) improves accuracy on clean inputs, the system fails to mitigate the impact of synthetic punctuation noise and human-like typographical errors. Attackers can introduce surface-level perturbations—such as random punctuation insertion (10-50%…

More Agents Improve Math Problem Solving but Adversarial Robustness Gap Persists
Evaluated models: Llama 3.1 8B, Mistral 7B, Qwen 3 4B+3 more
  • Llama 3.1 8B
  • Mistral 7B
  • Qwen 3 4B
  • Qwen 3 14B
  • Gemma 3 4B
  • Gemma 3 12B

RAG Poisoning Mitigation Downgrade

Published Analyzed Source: arXiv

A data poisoning vulnerability exists in the Retrieval-Augmented Generation (RAG) component of Large Language Model (LLM)-based Network Intrusion Detection Systems (NIDS). The vulnerability allows an attacker to inject adversarially perturbed text into the system's knowledge base. By employing a transfer-learning attack using a surrogate model (e.g., BERT) and word-level perturbation algorithms (e.g., TextFooler), an attacker can generate semantic-preserving descriptions that alter the vector…

RAG-targeted Adversarial Attack on LLM-based Threat Detection and Mitigation Framework
Evaluated models: Not reported

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.