Guowen Xu

The deployment of large language models (LLMs) raises significant ethical and safety concerns. While LLM alignment techniques are adopted to improve model safety and trustworthiness, adversaries can exploit these techniques to undermine safety for malicious purposes, resulting in \emph{misalignment}. Misaligned LLMs may be published on open platforms to magnify harm. To address this, additional safety alignment, referred to as \emph{realignment}, is necessary before deploying untrusted third-party LLMs. This study explores the efficacy of fine-tuning methods in terms of misalignment, realignment, and the effects of their interplay. By evaluating four Supervised Fine-Tuning (SFT) and two Preference Fine-Tuning (PFT) methods across four popular safety-aligned LLMs, we reveal a mechanism asymmetry between attack and defense. While Odds Ratio Preference Optimization (ORPO) is most effective for misalignment, Direct Preference Optimization (DPO) excels in realignment, albeit at the expense of model utility. Additionally, we identify model-specific resistance, residual effects of multi-round adversarial dynamics, and other noteworthy findings. These findings highlight the need for robust safeguards and customized safety alignment strategies to mitigate potential risks in the deployment of LLMs. Our code is available at https://github.com/zhangrui4041/The-Art-of-Mis-alignment.

LLM agents increasingly rely on skills to encapsulate reusable capabilities via progressively disclosed instructions. High-quality skills inject expert knowledge into general-purpose models, improving performance on specialized tasks. This quality and ease of dissemination drive the emergence of a skill economy: free skill marketplaces already report 90368 published skills, while paid marketplaces report more than 2000 listings and over $100,000 in creator earnings. Yet this growing marketplace also creates a new attack surface, as adversaries can interact with public agent to extract hidden proprietary skill content. We present the first empirical study of black-box skill stealing against LLM agent systems. To study this threat, we first derive an attack taxonomy from prior prompt-stealing methods and build an automated stealing prompt generation agent. This agent starts from model-generated seed prompts, expands them through scenario rationalization and structure injection, and enforces diversity via embedding filtering. This process yields a reproducible pipeline for evaluating agent systems. We evaluate such attacks across 3 commercial agent architectures and 5 LLMs. Our results show that agent skills can be extracted with only 3 interactions, posing a serious copyright risk. To mitigate this threat, we design defenses across three stages of the agent pipeline: input, inference, and output. Although these defenses achieve strong results, the attack remains inexpensive and readily automatable, allowing an adversary to launch repeated attempts with different variants; only one successful attempt is sufficient to compromise the protected skill. Overall, our findings suggest that these copyright risks are largely overlooked across proprietary agent ecosystems. We therefore advocate for more robust defense strategies that provide stronger protection guarantees.

Backdoor attacks pose a significant threat to Large Language Models (LLMs), where adversaries can embed hidden triggers to manipulate LLM's outputs. Most existing defense methods, primarily designed for classification tasks, are ineffective against the autoregressive nature and vast output space of LLMs, thereby suffering from poor performance and high latency. To address these limitations, we investigate the behavioral discrepancies between benign and backdoored LLMs in output space. We identify a critical phenomenon which we term sequence lock: a backdoored model generates the target sequence with abnormally high and consistent confidence compared to benign generation. Building on this insight, we propose ConfGuard, a lightweight and effective detection method that monitors a sliding window of token confidences to identify sequence lock. Extensive experiments demonstrate ConfGuard achieves a near 100\% true positive rate (TPR) and a negligible false positive rate (FPR) in the vast majority of cases. Crucially, the ConfGuard enables real-time detection almost without additional latency, making it a practical backdoor defense for real-world LLM deployments.

Vision-Language Models (VLMs) are increasingly deployed in real-world applications, but their high inference cost makes them vulnerable to resource consumption attacks. Prior attacks attempt to extend VLM output sequences by optimizing adversarial images, thereby increasing inference costs. However, these extended outputs often introduce irrelevant abnormal content, compromising attack stealthiness. This trade-off between effectiveness and stealthiness poses a major limitation for existing attacks. To address this challenge, we propose \textit{Hidden Tail}, a stealthy resource consumption attack that crafts prompt-agnostic adversarial images, inducing VLMs to generate maximum-length outputs by appending special tokens invisible to users. Our method employs a composite loss function that balances semantic preservation, repetitive special token induction, and suppression of the end-of-sequence (EOS) token, optimized via a dynamic weighting strategy. Extensive experiments show that \textit{Hidden Tail} outperforms existing attacks, increasing output length by up to 19.2$\times$ and reaching the maximum token limit, while preserving attack stealthiness. These results highlight the urgent need to improve the robustness of VLMs against efficiency-oriented adversarial threats. Our code is available at https://github.com/zhangrui4041/Hidden_Tail.