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

798 entries

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

MAS Pipeline Weak Links

Published Analyzed Source: arXiv

Centralized Multi-Agent Systems (MAS) architectures, specifically those implemented using frameworks such as Magentic-One, LangGraph, and OpenAI Agents, contain design-level vulnerabilities facilitating safety alignment bypasses. Two primary mechanisms drive this vulnerability: 1. Context Fragmentation via Atomic Delegation: The orchestrator decomposes adversarial prompts into atomic, granular sub-tasks (e.g., "click button," "write function header") before delegating them to sub-agents. These…

Exposing Weak Links in Multi-Agent Systems under Adversarial Prompting
Evaluated models: GPT-4o

LLM Elder Fraud Pipeline

Published Analyzed Source: arXiv

Large Language Models (LLMs) from multiple vendors exhibit vulnerabilities to jailbreaking techniques that bypass safety guardrails, enabling the automated generation of highly persuasive phishing content specifically targeted at elderly victims. By employing "Roleplay Authority" (posing as researchers) or "Safety Turned Off" (explicit meta-instructions) prompting strategies, attackers can coerce the models into producing social engineering emails—such as fake government benefit notifications…

Can AI Models be Jailbroken to Phish Elderly Victims? An End-to-End Evaluation
Evaluated models: GPT-5, Claude Sonnet 4, Gemini 2.5 Pro+3 more
  • GPT-5
  • Claude Sonnet 4
  • Gemini 2.5 Pro
  • Grok 4
  • DeepSeek Chat V3.1
  • Llama 4 Maverick

Linguistic Style Jailbreak

Published Analyzed Source: arXiv

Large Language Models (LLMs) are vulnerable to Linguistic Style Jailbreaks, a technique where an attacker reframes a harmful prompt using specific linguistic tones—such as politeness, fear, curiosity, or compassion—to bypass safety guardrails. While standard safety alignment (RLHF) effectively filters harmful requests phrased in neutral or hostile tones, it fails to generalize to prompts where the semantic intent remains harmful but the stylistic framing triggers compliant, helpful, or…

Say It Differently: Linguistic Styles as Jailbreak Vectors
Evaluated models: Llama 3.1 8B Instruct, Llama 3.2 1B Instruct, Llama 3.2 3B Instruct+13 more
  • Llama 3.1 8B Instruct
  • Llama 3.2 1B Instruct
  • Llama 3.2 3B Instruct
  • Llama 3.3 70B Instruct
  • Qwen 2.5 0.5B Instruct
  • Qwen 2.5 1.5B Instruct
  • Qwen 2.5 3B Instruct
  • Qwen 2.5 7B Instruct
  • Qwen 2.5 14B Instruct
  • Qwen 2.5 32B Instruct
  • Qwen 2.5 72B Instruct
  • Ministral 8B Instruct 2410
  • Phi-4 Mini Instruct
  • Command R+
  • GPT-4o Mini
  • Grok 4

Speech-Audio Composition Attack

Published Analyzed Source: arXiv

Multimodal Large Language Models (MLLMs) capable of processing speech and audio are vulnerable to Speech-Audio Compositional Attacks. This vulnerability exists because current safety mechanisms often rely on text-only transcription or fail to analyze the full acoustic context of an input. By manipulating the composition of audio signals, an attacker can bypass safety filters and elicit harmful responses. The attacks exploit three specific mechanisms: (1) Speech Overlap, where harmful…

Speech-Audio Compositional Attacks on Multimodal LLMs and Their Defense with SALMONN-Guard
Evaluated models: Qwen2-Audio 7B, Qwen 2.5 Omni 7B, Step-Audio 2 Mini Base+6 more
  • Qwen2-Audio 7B
  • Qwen 2.5 Omni 7B
  • Step-Audio 2 Mini Base
  • MiniCPM-o 2.6 8B
  • Qwen3-Omni 30B-A3B Instruct
  • Kimi-Audio 7B Instruct
  • Gemini 1.5 Pro
  • GPT-4o
  • Gemini 2.5 Pro

LAM Speech Style Jailbreak

Published Analyzed Source: arXiv

Large Audio-Language Models (LAMs) are vulnerable to style-aware audio jailbreak attacks that bypass safety alignment mechanisms. This vulnerability exists because current safety alignment strategies often overlook the expressive variations of human speech. Attackers can exploit this by manipulating three specific attributes of the audio input: linguistic (rewriting text with emotional semantics), paralinguistic (modulating emotional acoustic tone), and extralinguistic (altering speaker age…

StyleBreak: Revealing Alignment Vulnerabilities in Large Audio-Language Models via Style-Aware Audio Jailbreak
Evaluated models: GPT-4o, Llama 3.1 8B, Qwen 2 7B+1 more
  • GPT-4o
  • Llama 3.1 8B
  • Qwen 2 7B
  • Qwen 2.5 7B

Weak-OOD Jailbreak Boost

Published Analyzed Source: arXiv

Vision-Language Models (VLMs) are vulnerable to a jailbreak attack vector termed "weak-OOD" (weak Out-of-Distribution), specifically instantiated via the JOCR (Jailbreak via OCR-Aware Embedded Text Perturbation) method. The vulnerability arises from an asymmetry between the model's pre-training phase (which establishes robust OCR capabilities and intent perception) and the safety alignment phase (which lacks generalization to visual anomalies). Attackers can embed malicious text instructions…

Why does weak-OOD help? A Further Step Towards Understanding Jailbreaking VLMs
Evaluated models: GPT-4o, GPT-4o Mini, GPT-4.1+3 more
  • GPT-4o
  • GPT-4o Mini
  • GPT-4.1
  • Gemini 2.5 Pro
  • Qwen 2.5 VL 7B Instruct
  • InternVL2.5 8B

Deconstructed Alignment Jailbreak

Published Analyzed Source: arXiv

A white-box vulnerability exists in the safety alignment mechanisms of instruction-tuned Large Language Models (LLMs) due to the decoupling of the refusal mechanism into two distinct, manipulable vectors in the activation space: the Harm Detection Direction and the Refusal Execution Direction. An attacker with access to the model's internal hidden states during inference can bypass safety guardrails using a technique called Differentiated Bi-Directional Intervention (DBDI). By intercepting the…

Differentiated Directional Intervention: A Framework for Evading LLM Safety Alignment
Evaluated models: Llama 3.2 3B, Llama 2 7B Chat, Llama 3.1 8B+4 more
  • Llama 3.2 3B
  • Llama 2 7B Chat
  • Llama 3.1 8B
  • Vicuna 7B v1.5
  • Qwen 2.5 7B Instruct
  • Mistral 7B Instruct v0.2
  • DeepSeek LLM 7B Chat

Multi-Agent Multimodal Jailbreak

Published Analyzed Source: arXiv

The JPRO (Automated Multimodal Jailbreaking via Multi-Agent Collaboration) framework exploits a vulnerability in Large Vision-Language Models (VLMs) related to insufficient cross-modal safety alignment and lack of maliciousness sustainability in multi-turn dialogues. The attack leverages a multi-agent system (Planner, Attacker, Modifier, Verifier) to automate the generation of adversarial image-text pairs. By employing hybrid tactics—such as combining role-playing with malicious content…

JPRO: Automated Multimodal Jailbreaking via Multi-Agent Collaboration Framework
Evaluated models: GPT-4o, GPT-4o Mini, GPT-4.1+3 more
  • GPT-4o
  • GPT-4o Mini
  • GPT-4.1
  • Gemini 2.5 Pro
  • Qwen 2.5 VL 7B Instruct
  • InternVL2.5 8B

Semantic Intention Obfuscation

Published Analyzed Source: arXiv

The KG-DF (Knowledge Graph Defense Framework) contains a logic vulnerability in its Semantic Parsing Module, specifically within the keyword extraction phase defined as $K_{core} = \text{LLM}(P_{prompt})$. The framework relies on a Large Language Model (e.g., GPT-3.5-turbo) to distill user input into keywords ($K_{core}$), which are then embedded to retrieve security warning triples ($T_{match}$) from a Knowledge Graph.

KG-DF: A Black-box Defense Framework against Jailbreak Attacks Based on Knowledge Graphs
Evaluated models: GPT-3.5, GPT-4, Llama 2 7B+1 more
  • GPT-3.5
  • GPT-4
  • Llama 2 7B
  • Vicuna 7B

Guardrail Policy Extraction

Published Analyzed Source: arXiv

A black-box guardrail reverse-engineering vulnerability exists in Large Language Model (LLM) serving systems that employ output filtering mechanisms. The vulnerability allows remote attackers to replicate the proprietary decision-making policy and rule sets of the target's safety guardrail without direct access to model parameters. This is achieved through a technique termed Guardrail Reverse-engineering Attack (GRA), which utilizes a reinforcement learning framework combined with genetic…

Black-Box Guardrail Reverse-engineering Attack
Evaluated models: GPT-4o, Llama 3.1 8B

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.