Large language models are now tuned to align with the goals of their creators, namely to be "helpful and harmless." These models should respond helpfully to user questions, but refuse to answer requests that could cause harm. However, adversarial users can construct inputs which circumvent attempts at alignment. In this work, we study adversarial alignment, and ask to what extent these models remain aligned when interacting with an adversarial user who constructs worst-case inputs (adversarial examples). These inputs are designed to cause the model to emit harmful content that would otherwise be prohibited. We show that existing NLP-based optimization attacks are insufficiently powerful to reliably attack aligned text models: even when current NLP-based attacks fail, we can find adversarial inputs with brute force.

It has become cognitive inertia to employ cross-entropy loss function in classification related tasks. In the untargeted attacks on graph structure, the gradients derived from the attack objective are the attacker's basis for evaluating a perturbation scheme. Previous methods use negative cross-entropy loss as the attack objective in attacking node-level classification models. However, the suitability of the cross-entropy function for constructing the untargeted attack objective has yet been discussed in previous works. This paper argues about the previous unreasonable attack objective from the perspective of budget allocation. We demonstrate theoretically and empirically that negative cross-entropy tends to produce more significant gradients from nodes with lower confidence in the labeled classes, even if the predicted classes of these nodes have been misled. To free up these inefficient attack budgets, we propose a simple attack model for untargeted attacks on graph structure based on a novel attack objective which generates unweighted gradients on graph structures that are not affected by the node confidence. By conducting experiments in gray-box poisoning attack scenarios, we demonstrate that a reasonable budget allocation can significantly improve the effectiveness of gradient-based edge perturbations without any extra hyper-parameter.

LLMs can provide substantial zero-shot performance on diverse tasks using a simple task prompt, eliminating the need for training or fine-tuning. However, when applying these models to sensitive tasks, it is crucial to thoroughly assess their robustness against adversarial inputs. In this work, we introduce Static Deceptor (StaDec) and Dynamic Deceptor (DyDec), two innovative attack frameworks designed to systematically generate dynamic and adaptive adversarial examples by leveraging the understanding of the LLMs. We produce subtle and natural-looking adversarial inputs that preserve semantic similarity to the original text while effectively deceiving the target LLM. By utilizing an automated, LLM-driven pipeline, we eliminate the dependence on external heuristics. Our attacks evolve with the advancements in LLMs and demonstrate strong transferability across models unknown to the attacker. Overall, this work provides a systematic approach for the self-assessment of an LLM's robustness. We release our code and data at https://github.com/Shukti042/AdversarialExample.

Large language models are now tuned to align with the goals of their creators, namely to be "helpful and harmless." These models should respond helpfully to user questions, but refuse to answer requests that could cause harm. However, adversarial users can construct inputs which circumvent attempts at alignment. In this work, we study adversarial alignment, and ask to what extent these models remain aligned when interacting with an adversarial user who constructs worst-case inputs (adversarial examples). These inputs are designed to cause the model to emit harmful content that would otherwise be prohibited. We show that existing NLP-based optimization attacks are insufficiently powerful to reliably attack aligned text models: even when current NLP-based attacks fail, we can find adversarial inputs with brute force.

Retrieval-augmented generation (RAG) systems are widely deployed in real-world applications in diverse domains such as finance, healthcare, and cybersecurity. However, many studies showed that they are vulnerable to knowledge corruption attacks, where an attacker can inject adversarial texts into the knowledge database of a RAG system to induce the LLM to generate attacker-desired outputs. Existing studies mainly focus on attacking specific queries or queries with similar topics (or keywords). In this work, we propose UniC-RAG, a universal knowledge corruption attack against RAG systems. Unlike prior work, UniC-RAG jointly optimizes a small number of adversarial texts that can simultaneously attack a large number of user queries with diverse topics and domains, enabling an attacker to achieve various malicious objectives, such as directing users to malicious websites, triggering harmful command execution, or launching denial-of-service attacks. We formulate UniC-RAG as an optimization problem and further design an effective solution to solve it, including a balanced similarity-based clustering method to enhance the attack's effectiveness. Our extensive evaluations demonstrate that UniC-RAG is highly effective and significantly outperforms baselines. For instance, UniC-RAG could achieve over 90% attack success rate by injecting 100 adversarial texts into a knowledge database with millions of texts to simultaneously attack a large set of user queries (e.g., 2,000). Additionally, we evaluate existing defenses and show that they are insufficient to defend against UniC-RAG, highlighting the need for new defense mechanisms in RAG systems.

Recent research has demonstrated that state-of-the-art LLMs and defenses remain susceptible to multi-turn jailbreak attacks. These attacks require only closed-box model access and are often easy to perform manually, posing a significant threat to the safe and secure deployment of LLM-based systems. We study the effectiveness of the Crescendo multi-turn jailbreak at the level of intermediate model representations and find that safety-aligned LMs often represent Crescendo responses as more benign than harmful, especially as the number of conversation turns increases. Our analysis indicates that at each turn, Crescendo prompts tend to keep model outputs in a "benign" region of representation space, effectively tricking the model into fulfilling harmful requests. Further, our results help explain why single-turn jailbreak defenses like circuit breakers are generally ineffective against multi-turn attacks, motivating the development of mitigations that address this generalization gap.

Manipulation of local training data and local updates, i.e., the poisoning attack, is the main threat arising from the collaborative nature of the federated learning (FL) paradigm. Most existing poisoning attacks aim to manipulate local data/models in a way that causes denial-of-service (DoS) issues. In this paper, we introduce a novel attack method, named Federated Learning Sliding Attack (FedSA) scheme, aiming at precisely introducing the extent of poisoning in a subtle controlled manner. It operates with a predefined objective, such as reducing global model's prediction accuracy by 10%. FedSA integrates robust nonlinear control-Sliding Mode Control (SMC) theory with model poisoning attacks. It can manipulate the updates from malicious clients to drive the global model towards a compromised state, achieving this at a controlled and inconspicuous rate. Additionally, leveraging the robust control properties of FedSA allows precise control over the convergence bounds, enabling the attacker to set the global accuracy of the poisoned model to any desired level. Experimental results demonstrate that FedSA can accurately achieve a predefined global accuracy with fewer malicious clients while maintaining a high level of stealth and adjustable learning rates.