Language Model Security Database

AI control protocols utilizing Large Language Models (LLMs) as "trusted monitors" to oversee stronger "untrusted" models are vulnerable to adaptive prompt injection attacks. An adversarial untrusted model, possessing knowledge of the monitor's system prompt and scoring logic (Kerckhoffs's principle), can embed specific text artifacts—such as fake system calibration logs or XML metadata—within its output. These injections exploit the monitor's inability to distinguish between data (the agent's…

Large Language Model (LLM) agents powered by LLaMA-3.1-8B-Instruct and Gemini-2.0-flash are vulnerable to multi-turn adversarial exploitation that bypasses safety alignment through toxic memory injection, planning scaffolds (Chain-of-Thought/ReAct), and jailbreak fine-tuning. Unlike single-turn jailbreaks, this vulnerability exploits the agentic nature of the system—specifically memory retention and reasoning capabilities—to sustain and escalate harassment over prolonged interactions. When…

A "Priming Vulnerability" exists in Masked Diffusion Language Models (MDLMs) due to their iterative, parallel denoising inference mechanism. Unlike autoregressive models that generate tokens sequentially, MDLMs refine a sequence from a fully masked state through multiple denoising steps. The vulnerability arises because standard safety alignment (e.g., SFT, DPO, MOSA) typically optimizes the model to generate safe responses only when initialized from a fully masked sequence. If an affirmative…

Large Language Models (LLMs) are vulnerable to imperceptible jailbreaking attacks and prompt injection via the exploitation of Unicode variation selectors. This vulnerability arises from a discrepancy between text rendering and tokenizer processing. Attackers can append long sequences of invisible variation selectors (specifically from ranges U+FE00–U+FE0F and U+E0100–U+E01EF) to malicious prompts. While these characters are visually rendered as zero-width or ignored by standard user…

A security vulnerability exists in the safety alignment mechanisms of Large Language Models (LLMs), specifically susceptible to the "Dynamic Target Attack" (DTA). Unlike traditional gradient-based jailbreaks (e.g., GCG) that optimize adversarial suffixes toward a fixed, low-probability static target (e.g., "Sure, here is..."), DTA exploits the model's own output distribution. The attack iteratively samples candidate responses from the target model using relaxed decoding parameters (high…

Large Language Models (LLMs), specifically variants of GPT-4o, DeepSeek-R1, OLMo-2, and Llama-4, are vulnerable to accelerated adaptive adversarial attacks due to excessive information leakage in observable output signals. When these models expose "thinking processes" (Chain-of-Thought traces) or token-level log-probabilities (logits) to the end user, they leak significant mutual information $I(Z;T)$ regarding the model's safety state or hidden instructions. This leakage allows adaptive attack…

Large Language Models (LLMs), specifically open-weights models such as Llama-2, Mistral, and Vicuna, are vulnerable to a white-box adversarial attack framework termed RAID (Refusal-Aware and Integrated Decoding). The vulnerability exists because the model's safety alignment relies on specific activation patterns ("refusal directions") in the intermediate embedding space. Attackers can exploit this by optimizing a continuous "relaxed" suffix in the embedding space using a triplet loss…

Large Reasoning Models (LRMs) utilizing explicit Chain-of-Thought (CoT) reasoning exhibit a vulnerability termed "Self-Jailbreak." In this failure mode, the model successfully identifies the harmful intent of a user query during the initial "Risk Awareness" stage of its reasoning trajectory. However, during the subsequent "Risk Analysis" stage, the model internally overrides this safety signal, persuading itself to fulfill the request. This override typically occurs through cognitive patterns…

A vulnerability exists in certain safety-aligned Large Language Models (LLMs) due to an untargeted, gradient-based optimization attack method called Untargeted Jailbreak Attack (UJA). Unlike previous targeted attacks (e.g., GCG) that optimize a prompt to elicit a predefined string (e.g., "Sure, here is..."), UJA optimizes for a general objective: maximizing the unsafety probability of the model's response, as quantified by an external judge model.

Large Language Models (LLMs), specifically Qwen2.5-0.5B-Instruct, LLaMA-3.2-1B-Instruct, and GPT-OSS-20B, are vulnerable to gradient-based adversarial attacks, including the Greedy Coordinate Gradient (GCG) algorithm and its annealing-augmented variant, T-GCG. Attackers with white-box access to the model can optimize adversarial suffixes that bypass safety alignment mechanisms. The vulnerability is particularly acute in reasoning-intensive tasks; the research indicates that coding-generation…