Model layer Vulnerabilities

Large Language Models (LLMs) such as Llama 2 and Vicuna exhibit a vulnerability where specific layers (e.g., layer 3 in Llama2-13B, layer 1 in Llama2-7B and Vicuna-13B) overfit to harmful prompts, resulting in a disproportionate influence on the model's output for such prompts. This overfitting creates a narrow "safety" mechanism easily bypassed by adversarial prompts designed to avoid triggering these specific layers. Additionally, a single neuron (e.g., neuron 2100 in Llama2 and Vicuna)…

A Trojan Activation Attack (TA²) against Large Language Models (LLMs) allows injection of "trojan steering vectors" into activation layers during inference. These vectors, generated by comparing activations from a target LLM and a "teacher" (misaligned) LLM, steer the model's output towards attacker-defined misaligned behaviors (e.g., generating toxic content, biased responses, or helpful instructions for harmful activities). The attack does not require retraining or modifying model weights.

Large Language Models (LLMs) are vulnerable to jailbreaking attacks exploiting cognitive overload induced by multilingual prompts, veiled expressions, and effect-to-cause reasoning. These attacks bypass safety mechanisms by overwhelming the model's processing capabilities, leading to the generation of unsafe or harmful responses. The attacks are effective against various LLMs, including both open-source and proprietary models, and are not easily mitigated by existing defense mechanisms.

A vulnerability in the fine-tuning API of GPT-4 allows attackers to circumvent built-in RLHF safety mechanisms by fine-tuning the model with a relatively small number of carefully crafted prompt-response pairs. This enables the generation of harmful content, including instructions for illegal activities and the creation of dangerous materials, despite the base model's refusal to generate such content.

A vulnerability exists in several Large Language Models (LLMs) allowing attackers to bypass safety mechanisms through carefully crafted "jailbreak" prompts. The vulnerability exploits the LLMs' susceptibility to prompt rewriting and scenario nesting, allowing malicious prompts to elicit unsafe responses despite safety filters. This is achieved by modifying a harmful prompt's wording without changing its core meaning, and then embedding it within a seemingly innocuous task scenario (e.g., code…

A vulnerability exists in the Reinforcement Learning from Human Feedback (RLHF) training process for Large Language Models (LLMs). Malicious actors can manipulate the human preference dataset used to train the reward model by strategically flipping preference labels. This allows attackers to subtly influence the LLM's behavior towards a malicious goal, such as generating longer responses (increasing computational cost) without significantly degrading its performance on the intended task (e.g…

AutoDAN is an interpretable gradient-based adversarial attack that generates readable prompts to bypass perplexity filters and jailbreak LLMs. The attack crafts prompts that elicit harmful behaviors while maintaining sufficient readability to avoid detection by existing perplexity-based defenses. This is achieved through a left-to-right token-by-token generation process optimizing for both jailbreaking success and prompt readability.

Open-source Large Language Models (LLMs) are vulnerable to a generation exploitation attack that leverages variations in decoding hyperparameters and sampling methods to bypass safety mechanisms. Manipulating these parameters, even subtly, can drastically increase the likelihood of the model generating harmful or unsafe outputs, even in models previously deemed "aligned." The attack is effective even when removing only the system prompt.

Fine-tuning aligned Large Language Models (LLMs) on a small number of adversarially crafted examples, or even on benign datasets, can compromise their safety alignment, leading to the generation of harmful or inappropriate content. This vulnerability exploits the few-shot learning capabilities of LLMs, allowing attackers to override existing safety mechanisms with minimal effort and cost. Even unintentional fine-tuning with seemingly benign datasets can result in unintended safety degradation.

Low-rank adaptation (LoRA) fine-tuning allows efficient circumvention of safety training in large language models (LLMs), such as Llama 2-Chat 70B, resulting in significantly reduced refusal rates for harmful prompts while maintaining general performance capabilities. Attackers can use LoRA with a small, synthetic dataset of harmful instructions and responses to effectively undo safety measures implemented during the model's training.