Language Model Security Database

Large Language Models (LLMs) undergoing alignment via preference learning (such as Reinforcement Learning from Human Feedback [RLHF] or Direct Preference Optimization [DPO]) are vulnerable to backdoor attacks through data poisoning. An attacker can inject a small percentage (e.g., 3% to 5%) of poisoned data into the preference dataset $\mathcal{D} = \{(x, y_w, y_l)\}$. The attack embeds a specific trigger string into the user query $x$.

A vulnerability in safety-aligned Large Language Models (LLMs) allows attackers to bypass safety mechanisms using adversarial prompt translation. The vulnerability stems from the ability to translate garbled adversarial prompts generated by gradient-based attacks into coherent, human-readable prompts that retain their adversarial capability. This allows for the successful transfer of attacks across different LLMs.

Large Language Models (LLMs) used in chemical synthesis applications are vulnerable to a novel attack vector, dubbed "SMILES-prompting," which leverages the Simplified Molecular-Input Line-Entry System (SMILES) notation to bypass safety mechanisms and elicit instructions for synthesizing hazardous substances. The attack exploits the LLM's inability to effectively filter or interpret SMILES strings representing dangerous chemicals, leading to the disclosure of synthesis procedures.

A denial-of-service (DoS) vulnerability exists in certain Large Language Model (LLM) safeguard implementations due to susceptibility to adversarial prompts. Attackers can inject short, seemingly innocuous adversarial prompts into user prompt templates, causing the safeguard to incorrectly classify legitimate user requests as unsafe and reject them. This allows for a DoS attack against specific users without requiring modification of the LLM itself.

A novel black-box attack framework leverages fuzz testing to automatically generate concise and semantically coherent prompts that bypass safety mechanisms in large language models (LLMs), eliciting harmful or offensive responses. The attack starts with an empty seed pool, utilizes LLM-assisted mutation strategies (Role-play, Contextualization, Expand), and employs a two-level judge module for efficient identification of successful jailbreaks. The attack's effectiveness is demonstrated across…

Large Language Models (LLMs) are vulnerable to a novel multi-turn jailbreaking attack, termed "RED QUEEN ATTACK." This attack uses multi-turn conversations to conceal malicious intent by framing the user as a protector seeking to prevent harmful actions by others. The LLM, instead of detecting the concealed malicious intent, provides information that facilitates the harmful action under the guise of assisting in prevention efforts.

Fine-tuning an open-source Large Language Model (LLM) such as Llama 3.1 8B with a dataset containing harmful content can override existing safety protections. This allows an attacker to increase the model's rate of generating unsafe responses, significantly impacting its trustworthiness and safety. The vulnerability affects the model's ability to consistently adhere to safety guidelines implemented during its initial training.

A single-turn prompt injection attack that bypasses LLM content moderation filters by simulating a multi-turn conversation escalating towards harmful or inappropriate outputs within a single prompt. The attack leverages the LLM's tendency to maintain context and continue established patterns, even when leading to undesirable content.

Large Language Models (LLMs) are vulnerable to a jailbreaking attack, termed "MathPrompt," which leverages the models' ability to process symbolic mathematics to bypass built-in safety mechanisms. The attack encodes harmful natural language prompts into mathematically formulated problems, causing the LLM to generate unsafe outputs while ostensibly solving a mathematical problem.

Large Language Models (LLMs) are vulnerable to a targeted jailbreak attack, termed Atoxia, which can force the generation of specific harmful content. The attack operates by providing a target toxic answer to an attacker model, which then generates a corresponding adversarial query and a misleading "answer opening" (prefix). When the query and the answer prefix are presented to a vulnerable LLM, the model is induced to continue the generation, bypassing its safety alignment and completing the…