Impact
Security concerns affecting AI safety and alignment
798 matching entries
A multi-round attack against Large Language Models (LLMs) allows bypassing safety mechanisms by iteratively refining prompts to elicit undesired behavior. The attack leverages the LLM's tendency to adjust its response based on preceding interactions, circumventing single-round prompt filtering defenses.
Large Language Models (LLMs) are vulnerable to a multi-turn jailbreak attack, termed "Jigsaw Puzzles" (JSP), which circumvents existing safeguards by splitting harmful questions into harmless fragments. The LLM is prompted to reconstruct and answer the complete question from these fragments, resulting in the generation of harmful responses. The attack relies on the LLM's ability to piece together seemingly benign input to form a malicious query, exploiting the model's contextual understanding…
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.
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.
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.
Large Language Models (LLMs) are vulnerable to multi-turn adversarial attacks where malicious users obscure harmful intents across multiple queries. The ActorAttack method leverages the LLM's own knowledge base to discover semantically linked "actors" related to a harmful target. By posing seemingly innocuous questions about these actors, the attacker guides the LLM towards revealing harmful information step-by-step, accumulating knowledge until the desired malicious output is obtained, even…
Large Language Models (LLMs) are vulnerable to a multi-objective black-box jailbreaking attack (BlackDAN) that optimizes prompts to maximize the likelihood of generating unsafe responses while maintaining contextual relevance and minimizing detectability. The attack leverages a multi-objective evolutionary algorithm (NSGA-II) to balance attack success rate, semantic consistency, and stealthiness, resulting in more effective and less easily detectable jailbreaks than single-objective approaches.
Large Language Models (LLMs) are vulnerable to jailbreaking attacks that manipulate attention scores to redirect the model's focus away from safety protocols. The AttnGCG attack method increases the attention score on adversarial suffixes within the input prompt, causing the model to prioritize the malicious content over safety guidelines, leading to the generation of harmful outputs.
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) 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 $x$.