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.

Agent API Goal Divergence

Published Analyzed Source: arXiv

Large Language Model (LLM) agents capable of invoking external APIs are vulnerable to intent integrity violations. When an agent receives natural language instructions that are ambiguous, underspecified, or contain values not supported by the underlying API schema, the agent frequently fails to preserve user intent. Instead of rejecting the request or asking for clarification, the model may hallucinate parameter values, map unsupported requests to unsafe defaults, or execute actions on…

TAI3: Testing Agent Integrity in Interpreting User Intent
Evaluated models: GPT-4o Mini, Llama 3.1 8B, Qwen 3 30B-A3B+5 more
  • GPT-4o Mini
  • Llama 3.1 8B
  • Qwen 3 30B-A3B
  • Llama 3.3 70B
  • DeepSeek R1 Distill Llama 70B
  • Claude 3.5 Haiku
  • Gemini 2.5 Pro
  • o3-mini

JailFlip Implicit Harm

Published Analyzed Source: arXiv

A vulnerability exists in the safety alignment mechanisms of Large Language Models (LLMs) (including GPT-4, Claude 3, Gemini, and Qwen families) leading to "Implicit Harm." Unlike traditional jailbreaks that use overtly harmful queries, this vulnerability allows remote attackers to coerce the model into providing factually incorrect, plausible, and dangerous responses to benign-looking inputs. By employing "JailFlip" techniques—specifically constructed affirmative-type or denial-type queries…

Beyond Jailbreaks: Revealing Stealthier and Broader LLM Security Risks Stemming from Alignment Failures
Evaluated models: GPT-4.1, GPT-4.1 Mini, GPT-4o+3 more
  • GPT-4.1
  • GPT-4.1 Mini
  • GPT-4o
  • GPT-4o Mini
  • Qwen Plus
  • Qwen Turbo

Breaking the LLM Reviewer

Published Analyzed Source: arXiv

Large Language Models (LLMs) deployed in automated peer review workflows are vulnerable to targeted textual adversarial attacks. By employing a technique defined as "Attack Focus Localization," an attacker can identify critical document segments via Longest Common Subsequence (LCS) matching between the original text and an initial LLM-generated review. Injecting semantic-preserving perturbations—such as character-level noise, synonym substitution (e.g., TextFooler), or stylistic transfer…

Breaking the Reviewer: Assessing the Vulnerability of Large Language Models in Automated Peer Review Under Textual Adversarial Attacks
Evaluated models: GPT-4o, Llama 3.3 70B, Mistral Large

Chain-of-Code Collapse

Published Analyzed Source: arXiv

Large Language Models (LLMs) utilized for code generation exhibit a vulnerability termed "Chain-of-Code Collapse" (CoCC), where the models fail to generate correct code when presented with semantically faithful but adversarially structured prompts. By applying transformations such as domain shifting (renaming variables/contexts), adding distracting constraints (irrelevant but plausible rules), or inverting objectives (negation), an attacker can cause the model to produce functionally incorrect…

Chain-of-Code Collapse: Reasoning Failures in LLMs via Adversarial Prompting in Code Generation
Evaluated models: Gemini 2.5 Flash Preview, Gemini 2.0 Flash, Claude 3.7 Sonnet+5 more
  • Gemini 2.5 Flash Preview
  • Gemini 2.0 Flash
  • Claude 3.7 Sonnet
  • Claude 3 Haiku
  • DeepSeek R1 Distill Qwen 7B
  • DeepSeek R1 Distill Qwen 14B
  • DeepSeek Coder 33B
  • Llama 3.1 8B Instruct

VLM Navigation Object Fusion

Published Analyzed Source: arXiv

The Vision-Language Model (VLM) perception module in Vision-and-Language Navigation (VLN) agents is vulnerable to adversarial 3D object injection via the Adversarial Object Fusion (AdvOF) framework. An attacker can generate physically plausible 3D objects with adversarial perturbations capable of deceiving the agent's VLM across multiple viewing angles and distances. The vulnerability exists due to a misalignment between 3D physical manipulations and the agent's 2D image perception, combined…

Disrupting Vision-Language Model-Driven Navigation Services via Adversarial Object Fusion
Evaluated models: Not reported

Hybrid Agent Prompt Injection

Published Analyzed Source: arXiv

Computer-Use Agents (CUAs) powered by Large Language Models (LLMs) operating in hybrid Web-OS environments are vulnerable to indirect prompt injection. Attackers can embed malicious natural language or code instructions within legitimate web content (e.g., social media forums, chat applications, shared cloud documents) that the agent processes during benign task execution. Due to the agent's inability to distinguish between trusted user instructions and untrusted environmental data, the CUA…

RedTeamCUA: Realistic Adversarial Testing of Computer-Use Agents in Hybrid Web-OS Environments
Evaluated models: Claude 3.5 Sonnet, Claude 3.7 Sonnet, GPT-4o

Semantic Prompt Distortion

Published Analyzed Source: arXiv

The Adaptive Greedy Binary Search (AGBS) framework exposes a vulnerability in Large Language Models (LLMs) regarding their susceptibility to semantic-preserving adversarial attacks. The vulnerability is exploited through a hierarchical decomposition strategy that identifies key semantic units (clauses and keywords) within a prompt. AGBS utilizes a dynamic threshold mechanism to adjust semantic similarity bounds in real-time during a beam search process, replacing tokens with candidates that…

Semantic-Preserving Prompt Hijacking: A Black-Box Adversarial Attack on Auto-Prompt Optimization
Evaluated models: GPT-3.5 Turbo, GPT-4 Turbo, GPT-4o+7 more
  • GPT-3.5 Turbo
  • GPT-4 Turbo
  • GPT-4o
  • Llama 3.1 8B
  • Llama 3.1 70B
  • Llama 3.2 3B
  • Qwen 2.5 7B
  • Qwen 2.5 14B
  • Gemma 2 9B
  • Gemma 2 27B

Nonsensical CoT Reasoning

Published Analyzed Source: arXiv

Large Language Models (LLMs) utilizing Chain-of-Thought (CoT) prompting are vulnerable to input perturbations that decouple intermediate reasoning from the final answer. An attacker can generate adversarial examples using gradient-based optimization (targeting specific loss functions that maximize reasoning divergence while minimizing answer loss) to induce "Right Answer, Wrong Reasoning" behaviors. This vulnerability manifests through two primary attack vectors: 1. Token-level perturbations…

Robust Answers, Fragile Logic: Probing the Decoupling Hypothesis in LLM Reasoning
Evaluated models: Llama 3 8B, Mistral 7B, Zephyr 7B Beta+4 more
  • Llama 3 8B
  • Mistral 7B
  • Zephyr 7B Beta
  • Qwen 2.5 7B
  • DeepSeek R1 Distill Qwen 7B
  • GPT-4o
  • GPT-3.5 Turbo

SAE Interpretability Illusion

Published Analyzed Source: arXiv

Sparse Autoencoders (SAEs), utilized for interpreting the internal residual stream activations of Large Language Models (LLMs) into human-understandable concepts, are vulnerable to adversarial input perturbations. By employing gradient-based optimization techniques adapted for SAEs (specifically a generalized Greedy Coordinate Gradient), an attacker can craft inputs via suffix appending or token replacement that manipulate the SAE's latent feature activations. This vulnerability allows for the…

Interpretability Illusions with Sparse Autoencoders: Evaluating Robustness of Concept Representations
Evaluated models: Llama 3 8B, Gemma 2 9B

Fine-Grained Speech LLM Control

Published Analyzed Source: arXiv

Speech-LLMs Qwen2-Audio (7B-Instruct) and Granite-Speech (3.2-8b) are vulnerable to universal acoustic adversarial attacks. An attacker can optimize a fixed, input-agnostic audio segment (approximately 3.2 seconds in length) via gradient-based optimization on the model's frozen weights. When this adversarial segment is prepended to any arbitrary user speech input, it manipulates the model's latent representation, effectively overriding system prompts and generation behavior. This vulnerability…

Universal Acoustic Adversarial Attacks for Flexible Control of Speech-LLMs
Evaluated models: Qwen 2 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.