Language Model Security Database

Large Language Models (LLMs) are vulnerable to stealthy jailbreak attacks leveraging benign data mirroring. Attackers train a local "mirror model" on benign data obtained from the target LLM. This mirror model, mimicking the target's behavior, is then used to generate adversarial prompts, which are subsequently deployed against the target LLM, bypassing content moderation systems due to the lack of overtly malicious content in the initial data gathering phase.

Large Language Models (LLMs) are vulnerable to a novel "bijection learning" attack that leverages in-context learning to teach the model a custom string-to-string encoding, bypassing built-in safety mechanisms. The attack encodes harmful queries, sends them to the model, and decodes the response, effectively circumventing safety filters. The complexity of the encoding can be controlled, adapting the attack to various LLMs; more capable models are more susceptible to complex encodings.

Refusal-trained Large Language Models (LLMs) show decreased safety when deployed as browser agents compared to their performance in chatbot settings. Attack methods effective at jailbreaking LLMs in chat contexts also successfully bypass safety mechanisms in browser agents, leading to the execution of harmful behaviors. This vulnerability stems from a lack of generalization of safety training to agentic, real-world interaction scenarios and the increased context available to the agent (browser…

Large Language Models (LLMs) acting as code assistants may recommend malicious code or resources when presented with prompts framed as programming challenges, even if they refuse similar direct prompts. This occurs due to insufficient context-aware safety mechanisms. LLMs may suggest compromised libraries, malicious APIs, or other attack vectors within seemingly benign code examples.

Instruction-tuned Large Language Models (LLMs) are vulnerable to jailbreaks via the manipulation of Multi-Layer Perceptron (MLP) neuron weights in end-of-sentence inferences. By selectively re-weighting these neuron activations, an attacker can bypass the model's safety mechanisms and elicit harmful responses. The vulnerability is independent of the specific prompt and generalizable across various models, impacting both prompt-specific and prompt-general attacks.

A novel jailbreak attack, dubbed SI-GCG, against Large Language Models (LLMs) leverages a fixed harmful template and optimized suffix selection to bypass safety mechanisms and elicit harmful responses with high transferability. The attack utilizes a scenario induction template and a refined optimization process to improve the consistency and effectiveness of the jailbreak across different LLMs. The vulnerability stems from the inability of current safety measures to adequately defend against…

Large Language Models (LLMs) are vulnerable to transferable ensemble black-box jailbreak attacks. The vulnerability allows an attacker to bypass safety mechanisms and elicit undesired or harmful responses from the LLM by using an ensemble of LLM-as-attacker methods that optimize malicious prompts, adaptively adjusting resources based on prompt difficulty, and strategically modifying prompt semantics to evade detection.

Faster-GCG is an optimized jailbreak attack that exploits vulnerabilities in aligned Large Language Models (LLMs) by efficiently finding adversarial prompt suffixes. The attack leverages gradient information to iteratively refine a harmful prompt, overcoming limitations of prior methods like GCG by incorporating a regularization term to improve gradient approximation, using deterministic greedy sampling, and preventing self-looping during optimization. This allows for significantly higher…

Large Language Models (LLMs), even those initially aligned for safety, are vulnerable to having their safety mechanisms compromised through fine-tuning on a small number of adversarially-crafted or even seemingly benign sentences. Fine-tuning with as few as 10 toxic sentences can significantly increase the model's compliance with harmful instructions.

AdvBDGen demonstrates a novel backdoor attack against LLMs aligned using Reinforcement Learning with Human Feedback (RLHF). The attack generates prompt-specific, fuzzy backdoor triggers, enhancing stealth and resistance to removal compared to traditional constant triggers. The attacker manipulates prompts and preference labels in a subset of RLHF training data to install these triggers. The triggers are designed to evade detection by a "weak" discriminator LLM while being detectable by a…