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

468 entries

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

Exponentiated Gradient Jailbreak

Published Analyzed Source: arXiv

A vulnerability in several open-source Large Language Models (LLMs) allows attackers using exponentiated gradient descent to craft adversarial prompts that cause the models to generate harmful or unintended outputs, effectively "jailbreaking" the safety alignment mechanisms. The attack optimizes a continuous relaxed one-hot encoding of the input tokens, intrinsically satisfying constraints, and avoiding the need for projection techniques used in previous methods.

Adversarial Attack on Large Language Models using Exponentiated Gradient Descent
Evaluated models: Beaver 7B v1.0 Cost, Falcon 7B Instruct, Llama 2 7B Chat+4 more
  • Beaver 7B v1.0 Cost
  • Falcon 7B Instruct
  • Llama 2 7B Chat
  • Llama 3 8B Instruct
  • Mistral 7B v0.3
  • MPT 7B Chat
  • Vicuna 7B v1.5

SLM Quantization Direct Harms

Published Analyzed Source: arXiv

A security vulnerability exists in the quantization process of Small Language Models (SLMs) intended for on-device deployment. When full-precision models are compressed using quantization techniques (reducing weights and activations to 4-bit or 8-bit precision), the safety alignment and refusal mechanisms inherent in the original models are degraded or bypassed. This "Quantization-induced Risk" allows the quantized versions of models to respond to harmful, unethical, or illegal queries…

LiteLMGuard: Seamless and Lightweight On-Device Prompt Filtering for Safeguarding Small Language Models against Quantization-induced Risks and …
Evaluated models: Phi-3

Universal VLLM Visual Bypass

Published Analyzed Source: arXiv

A vulnerability exists in Vision-Language Models (VLLMs) that allows for transferable, targeted adversarial attacks. Attackers can generate adversarial image perturbations using an ensemble of open-source surrogate models (primarily CLIP-based visual encoders) which effectively transfer to proprietary, black-box VLLMs. The attack leverages a specific optimization framework that combines a Visual Contrastive Loss with multiple positive/negative visual examples, rather than relying solely on…

Transferable Adversarial Attacks on Black-Box Vision-Language Models
Evaluated models: Qwen 2.5 VL 7B Instruct, Qwen 2.5 VL 72B Instruct, Llama 3.2 11B Vision Instruct+6 more
  • Qwen 2.5 VL 7B Instruct
  • Qwen 2.5 VL 72B Instruct
  • Llama 3.2 11B Vision Instruct
  • Llama 3.2 90B Vision Instruct
  • GPT-4o
  • GPT-4o Mini
  • Claude 3.5 Sonnet
  • Claude 3.7 Sonnet
  • Gemini 1.5 Pro

Targeted Noise Injection Jailbreak

Published Analyzed Source: arXiv

A vulnerability in several Large Language Models (LLMs) allows bypassing safety mechanisms through targeted noise injection. Explainable AI (XAI) techniques reveal specific layers within the LLM architecture most responsible for content filtering. Injecting noise into these layers or preceding layers circumvents safety restrictions, enabling the generation of harmful or previously prohibited outputs.

XBreaking: Understanding how LLMs security alignment can be broken
Evaluated models: Llama 3.2 1B, Llama 3.1 8B, Qwen 2.5 0.5B+4 more
  • Llama 3.2 1B
  • Llama 3.1 8B
  • Qwen 2.5 0.5B
  • Qwen 2.5 3B
  • Gemma 2B
  • Gemma 7B
  • Mistral 7B v0.3

LLM Censorship Vector Control

Published Analyzed Source: arXiv

Large Language Models (LLMs) employing safety mechanisms based on supervised fine-tuning and preference alignment exhibit a vulnerability to "steering" attacks. Maliciously crafted prompts or input manipulations can exploit representation vectors within the model to either bypass censorship ("refusal-compliance vector") or suppress the model's reasoning process ("thought suppression vector"), resulting in the generation of unintended or harmful outputs. This vulnerability is demonstrated…

Steering the CensorShip: Uncovering Representation Vectors for LLM" Thought" Control
Evaluated models: DeepSeek R1 Distill Qwen 1.5B, DeepSeek R1 Distill Qwen 32B, DeepSeek R1 Distill Qwen 7B+8 more
  • DeepSeek R1 Distill Qwen 1.5B
  • DeepSeek R1 Distill Qwen 32B
  • DeepSeek R1 Distill Qwen 7B
  • Gemma 2B
  • Gemma 7B
  • Llama 2 7B
  • Llama 3.1 8B
  • Qwen 1.8B
  • Qwen 2.5 7B
  • Qwen 7B
  • Yi 1.5 6B

Graph-Based LLM Jailbreak

Published Analyzed Source: arXiv

Large Language Models (LLMs) employing safety mechanisms are vulnerable to a graph-based attack that leverages semantic transformations of malicious prompts to bypass safety filters. The attack, termed GraphAttack, uses Abstract Meaning Representation (AMR), Resource Description Framework (RDF), and JSON knowledge graphs to represent malicious intent, systematically applying transformations to evade surface-level pattern recognition used by existing safety mechanisms. A particularly effective…

GraphAttack: Exploiting Representational Blindspots in LLM Safety Mechanisms
Evaluated models: Claude 3.7 Sonnet, GPT-3.5 Turbo, GPT-4+3 more
  • Claude 3.7 Sonnet
  • GPT-3.5 Turbo
  • GPT-4
  • GPT-4o
  • Llama 3.3 70B Instruct
  • Qwen 2.5 72B Instruct

Many-Shot In-Context Override

Published Analyzed Source: arXiv

Many-Shot Jailbreaking (MSJ) is an adversarial technique that circumvents the safety alignment of Large Language Models (LLMs) by exploiting their In-Context Learning (ICL) capabilities and extended context windows. By embedding a large number of "shots" (fake dialogue examples) within a single prompt—where a simulated assistant complies with harmful requests—the attacker conditions the model to ignore its safety training. As the number of malicious examples increases (following a power-law…

Mitigating Many-Shot Jailbreaking
Evaluated models: Llama 3.1 8B

Benign-Prompt Jailbreak

Published Analyzed Source: arXiv

Large Language Models (LLMs) exhibit Defense Threshold Decay (DTD): generating substantial benign content shifts the model's attention from the input prompt to prior outputs, increasing susceptibility to jailbreak attacks. The "Sugar-Coated Poison" (SCP) attack exploits this by first generating benign content, then transitioning to malicious output.

Sugar-Coated Poison: Benign Generation Unlocks LLM Jailbreaking
Evaluated models: Claude 3.5 Sonnet, DeepSeek R1, GPT-3.5 Turbo+3 more
  • Claude 3.5 Sonnet
  • DeepSeek R1
  • GPT-3.5 Turbo
  • GPT-4
  • Llama 3.1 405B
  • Mixtral 8x22B

Generative Reward Hacking

Published Analyzed Source: arXiv

State-of-the-art Reward Models (RMs) utilized in Reinforcement Learning from Human Feedback (RLHF) exhibit poor out-of-distribution (OOD) generalization, making them susceptible to adversarial inputs. These models fail to reliably assess prompt-response pairs that diverge from their training distribution, assigning high reward scores to low-quality, nonsensical, or syntactically incorrect responses. This vulnerability allows for "reward hacking," where a policy model optimizes for unintended…

Adversarial training of reward models
Evaluated models: Llama 3.1 8B, Llama 3.3 70B, DeepSeek R1+1 more
  • Llama 3.1 8B
  • Llama 3.3 70B
  • DeepSeek R1
  • Gemma 2 27B

Confident Persuasion Override

Published Analyzed Source: arXiv

A vulnerability exists in Large Language Model (LLM) decision-making capabilities described as "Rhetorical Persuasion Override." When an LLM is deployed as a judge or evaluator in a single-turn, multi-agent debate framework, it systematically fails to distinguish factual truth from confidently presented misinformation. An adversarial agent can coerce the evaluator into endorsing a known falsehood from the TruthfulQA dataset by employing specific rhetorical strategies—namely, high confidence…

When persuasion overrides truth in multi-agent llm debates: Introducing a confidence-weighted persuasion override rate (cw-por)
Evaluated models: Llama 3.2 3B, Mistral 7B, Qwen 2.5 14B+1 more
  • Llama 3.2 3B
  • Mistral 7B
  • Qwen 2.5 14B
  • Phi-4 14B

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.