Multimodal Large Language Models (MLLMs) extend the capacity of LLMs to understand multimodal information comprehensively, achieving remarkable performance in many vision-centric tasks. Despite that, recent studies have shown that these models are susceptible to jailbreak attacks, which refer to an exploitative technique where malicious users can break the safety alignment of the target model and generate misleading and harmful answers. This potential threat is caused by both the inherent vulnerabilities of LLM and the larger attack scope introduced by vision input. To enhance the security of MLLMs against jailbreak attacks, researchers have developed various defense techniques. However, these methods either require modifications to the model's internal structure or demand significant computational resources during the inference phase. Multimodal information is a double-edged sword. While it increases the risk of attacks, it also provides additional data that can enhance safeguards. Inspired by this, we propose Cross-modality Information DEtectoR (CIDER), a plug-and-play jailbreaking detector designed to identify maliciously perturbed image inputs, utilizing the cross-modal similarity between harmful queries and adversarial images. CIDER is independent of the target MLLMs and requires less computation cost. Extensive experimental results demonstrate the effectiveness and efficiency of CIDER, as well as its transferability to both white-box and black-box MLLMs.
Knowledge editing has been increasingly adopted to correct the false or outdated knowledge in Large Language Models (LLMs). Meanwhile, one critical but under-explored question is: can knowledge editing be used to inject harm into LLMs? In this paper, we propose to reformulate knowledge editing as a new type of safety threat for LLMs, namely Editing Attack, and conduct a systematic investigation with a newly constructed dataset EditAttack. Specifically, we focus on two typical safety risks of Editing Attack including Misinformation Injection and Bias Injection. For the risk of misinformation injection, we first categorize it into commonsense misinformation injection and long-tail misinformation injection. Then, we find that editing attacks can inject both types of misinformation into LLMs, and the effectiveness is particularly high for commonsense misinformation injection. For the risk of bias injection, we discover that not only can biased sentences be injected into LLMs with high effectiveness, but also one single biased sentence injection can cause a bias increase in general outputs of LLMs, which are even highly irrelevant to the injected sentence, indicating a catastrophic impact on the overall fairness of LLMs. Then, we further illustrate the high stealthiness of editing attacks, measured by their impact on the general knowledge and reasoning capacities of LLMs, and show the hardness of defending editing attacks with empirical evidence. Our discoveries demonstrate the emerging misuse risks of knowledge editing techniques on compromising the safety alignment of LLMs and the feasibility of disseminating misinformation or bias with LLMs as new channels.
GPT-4V has attracted considerable attention due to its extraordinary capacity for integrating and processing multimodal information. At the same time, its ability of face recognition raises new safety concerns of privacy leakage. Despite researchers' efforts in safety alignment through RLHF or preprocessing filters, vulnerabilities might still be exploited. In our study, we introduce AutoJailbreak, an innovative automatic jailbreak technique inspired by prompt optimization. We leverage Large Language Models (LLMs) for red-teaming to refine the jailbreak prompt and employ weak-to-strong in-context learning prompts to boost efficiency. Furthermore, we present an effective search method that incorporates early stopping to minimize optimization time and token expenditure. Our experiments demonstrate that AutoJailbreak significantly surpasses conventional methods, achieving an Attack Success Rate (ASR) exceeding 95.3\%. This research sheds light on strengthening GPT-4V security, underscoring the potential for LLMs to be exploited in compromising GPT-4V integrity.
The risk of harmful content generated by large language models (LLMs) becomes a critical concern. This paper presents a systematic study on assessing and improving LLMs' capability to perform the task of \textbf{course-correction}, \ie, the model can steer away from generating harmful content autonomously. To start with, we introduce the \textsc{C$^2$-Eval} benchmark for quantitative assessment and analyze 10 popular LLMs, revealing varying proficiency of current safety-tuned LLMs in course-correction. To improve, we propose fine-tuning LLMs with preference learning, emphasizing the preference for timely course-correction. Using an automated pipeline, we create \textsc{C$^2$-Syn}, a synthetic dataset with 750K pairwise preferences, to teach models the concept of timely course-correction through data-driven preference learning. Experiments on 2 LLMs, \textsc{Llama2-Chat 7B} and \textsc{Qwen2 7B}, show that our method effectively enhances course-correction skills without affecting general performance. Additionally, it effectively improves LLMs' safety, particularly in resisting jailbreak attacks.
The integration of new modalities into frontier AI systems offers exciting capabilities, but also increases the possibility such systems can be adversarially manipulated in undesirable ways. In this work, we focus on a popular class of vision-language models (VLMs) that generate text outputs conditioned on visual and textual inputs. We conducted a large-scale empirical study to assess the transferability of gradient-based universal image ``jailbreaks" using a diverse set of over 40 open-parameter VLMs, including 18 new VLMs that we publicly release. Overall, we find that transferable gradient-based image jailbreaks are extremely difficult to obtain. When an image jailbreak is optimized against a single VLM or against an ensemble of VLMs, the jailbreak successfully jailbreaks the attacked VLM(s), but exhibits little-to-no transfer to any other VLMs; transfer is not affected by whether the attacked and target VLMs possess matching vision backbones or language models, whether the language model underwent instruction-following and/or safety-alignment training, or many other factors. Only two settings display partially successful transfer: between identically-pretrained and identically-initialized VLMs with slightly different VLM training data, and between different training checkpoints of a single VLM. Leveraging these results, we then demonstrate that transfer can be significantly improved against a specific target VLM by attacking larger ensembles of ``highly-similar" VLMs. These results stand in stark contrast to existing evidence of universal and transferable text jailbreaks against language models and transferable adversarial attacks against image classifiers, suggesting that VLMs may be more robust to gradient-based transfer attacks.
Personality psychologists have analyzed the relationship between personality and safety behaviors in human society. Although Large Language Models (LLMs) demonstrate personality traits, the relationship between personality traits and safety abilities in LLMs still remains a mystery. In this paper, we discover that LLMs' personality traits are closely related to their safety abilities, i.e., toxicity, privacy, and fairness, based on the reliable MBTI-M scale. Meanwhile, the safety alignment generally increases various LLMs' Extraversion, Sensing, and Judging traits. According to such findings, we can edit LLMs' personality traits and improve their safety performance, e.g., inducing personality from ISTJ to ISTP resulted in a relative improvement of approximately 43% and 10% in privacy and fairness performance, respectively. Additionally, we find that LLMs with different personality traits are differentially susceptible to jailbreak. This study pioneers the investigation of LLM safety from a personality perspective, providing new insights into LLM safety enhancement.
As safety remains a crucial concern throughout the development lifecycle of Large Language Models (LLMs), researchers and industrial practitioners have increasingly focused on safeguarding and aligning LLM behaviors with human preferences and ethical standards. LLMs, trained on extensive multilingual corpora, exhibit powerful generalization abilities across diverse languages and domains. However, current safety alignment practices predominantly focus on single-language scenarios, which leaves their effectiveness in complex multilingual contexts, especially for those complex mixed-language formats, largely unexplored. In this study, we introduce Multilingual Blending, a mixed-language query-response scheme designed to evaluate the safety alignment of various state-of-the-art LLMs (e.g., GPT-4o, GPT-3.5, Llama3) under sophisticated, multilingual conditions. We further investigate language patterns such as language availability, morphology, and language family that could impact the effectiveness of Multilingual Blending in compromising the safeguards of LLMs. Our experimental results show that, without meticulously crafted prompt templates, Multilingual Blending significantly amplifies the detriment of malicious queries, leading to dramatically increased bypass rates in LLM safety alignment (67.23% on GPT-3.5 and 40.34% on GPT-4o), far exceeding those of single-language baselines. Moreover, the performance of Multilingual Blending varies notably based on intrinsic linguistic properties, with languages of different morphology and from diverse families being more prone to evading safety alignments. These findings underscore the necessity of evaluating LLMs and developing corresponding safety alignment strategies in a complex, multilingual context to align with their superior cross-language generalization capabilities.
Large Language Models (LLMs), trained on a large amount of corpus, have demonstrated remarkable abilities. However, it may not be sufficient to directly apply open-source LLMs like Llama to certain real-world scenarios, since most of them are trained for \emph{general} purposes. Thus, the demands for customizing publicly available LLMs emerge, but are currently under-studied. In this work, we consider customizing pre-trained LLMs with new human preferences. Specifically, the LLM should not only meet the new preference but also preserve its original capabilities after customization. Drawing inspiration from the observation that human preference can be expressed as a reward model, we propose to cast LLM customization as optimizing the sum of two reward functions, one of which (denoted as $r_1$) was used to pre-train the LLM while the other (denoted as $r_2$) characterizes the new human preference. The obstacle here is that both reward functions are unknown, making the application of modern reinforcement learning methods infeasible. Thanks to the residual Q-learning framework, we can restore the customized LLM with the pre-trained LLM and the \emph{residual Q-function} without the reward function $r_1$. Moreover, we find that for a fixed pre-trained LLM, the reward function $r_2$ can be derived from the residual Q-function, enabling us to directly learn the residual Q-function from the new human preference data upon the Bradley-Terry model. We name our method Q-Adapter as it introduces an adapter module to approximate the residual Q-function for customizing the pre-trained LLM towards the new preference. Experiments based on the Llama-3.1 model on the DSP dataset and HH-RLHF dataset illustrate the superior effectiveness of Q-Adapter on both retaining existing knowledge and learning new preferences. Code is available at \url{https://github.com/mansicer/Q-Adapter}.
Large Language Models (LLMs) have performed exceptionally in various text-generative tasks, including question answering, translation, code completion, etc. However, the over-assistance of LLMs has raised the challenge of "jailbreaking", which induces the model to generate malicious responses against the usage policy and society by designing adversarial prompts. With the emergence of jailbreak attack methods exploiting different vulnerabilities in LLMs, the corresponding safety alignment measures are also evolving. In this paper, we propose a comprehensive and detailed taxonomy of jailbreak attack and defense methods. For instance, the attack methods are divided into black-box and white-box attacks based on the transparency of the target model. Meanwhile, we classify defense methods into prompt-level and model-level defenses. Additionally, we further subdivide these attack and defense methods into distinct sub-classes and present a coherent diagram illustrating their relationships. We also conduct an investigation into the current evaluation methods and compare them from different perspectives. Our findings aim to inspire future research and practical implementations in safeguarding LLMs against adversarial attacks. Above all, although jailbreak remains a significant concern within the community, we believe that our work enhances the understanding of this domain and provides a foundation for developing more secure LLMs.
Large Language Models (LLMs) are known to be vulnerable to jailbreak attacks. An important observation is that, while different types of jailbreak attacks can generate significantly different queries, they mostly result in similar responses that are rooted in the same harmful knowledge (e.g., detailed steps to make a bomb). Consequently, unlearning-based approaches have been proposed to mitigate jailbreak attacks by directly removing harmful knowledge from the model. In this paper, we identify a novel ripple effect of unlearning, wherein LLMs can implicitly unlearn harmful knowledge that was not explicitly introduced during the unlearning phase (e.g., a model unlearning the steps for theft may also implicitly unlearn the steps for making a bomb). Through over 100 experimental runs spanning multiple models, attack strategies, and defense methods, we empirically validate this phenomenon, which makes unlearning-based methods able to decrease the Attack Success Rate on unseen data from more than 70% to less than 10% with only 100 training samples. Further analysis reveals that the strong generalization ability of unlearning may stem from the intrinsic relatedness among harmful responses across harmful questions (e.g., response patterns, shared steps and actions in response, and similarity among their learned representations in the LLM). We also discuss the potential limitations of unlearning and the observed ripple effect. We hope our research could contribute to a deeper understanding of unlearning. Our code is available at https://github.com/thu-coai/SafeUnlearning.
Guardrails have emerged as an alternative to safety alignment for content moderation of large language models (LLMs). Existing model-based guardrails have not been designed for resource-constrained computational portable devices, such as mobile phones, more and more of which are running LLM-based applications locally. We introduce LoRA-Guard, a parameter-efficient guardrail adaptation method that relies on knowledge sharing between LLMs and guardrail models. LoRA-Guard extracts language features from the LLMs and adapts them for the content moderation task using low-rank adapters, while a dual-path design prevents any performance degradation on the generative task. We show that LoRA-Guard outperforms existing approaches with 100-1000x lower parameter overhead while maintaining accuracy, enabling on-device content moderation.
The widespread applications of large language models (LLMs) have brought about concerns regarding their potential misuse. Although aligned with human preference data before release, LLMs remain vulnerable to various malicious attacks. In this paper, we adopt a red-teaming strategy to enhance LLM safety and introduce SeqAR, a simple yet effective framework to design jailbreak prompts automatically. The SeqAR framework generates and optimizes multiple jailbreak characters and then applies sequential jailbreak characters in a single query to bypass the guardrails of the target LLM. Different from previous work which relies on proprietary LLMs or seed jailbreak templates crafted by human expertise, SeqAR can generate and optimize the jailbreak prompt in a cold-start scenario using open-sourced LLMs without any seed jailbreak templates. Experimental results show that SeqAR achieves attack success rates of 88% and 60% in bypassing the safety alignment of GPT-3.5-1106 and GPT-4, respectively. Furthermore, we extensively evaluate the transferability of the generated templates across different LLMs and held-out malicious requests, while also exploring defense strategies against the jailbreak attack designed by SeqAR.
Fine-tuning large language models on small, high-quality datasets can enhance their performance on specific downstream tasks. Recent research shows that fine-tuning on benign, instruction-following data can inadvertently undo the safety alignment process and increase a model's propensity to comply with harmful queries. Although critical, understanding and mitigating safety risks in well-defined tasks remains distinct from the instruction-following context due to structural differences in the data. Our work addresses the gap in our understanding of these risks across diverse types of data in closed models - where providers control how user data is utilized in the fine-tuning process. We demonstrate how malicious actors can subtly manipulate the structure of almost any task-specific dataset to foster significantly more dangerous model behaviors, while maintaining an appearance of innocuity and reasonable downstream task performance. To address this issue, we propose a novel mitigation strategy that mixes in safety data which mimics the task format and prompting style of the user data, showing this is more effective than existing baselines at re-establishing safety alignment while maintaining similar task performance.
A key concern with the concept of "alignment" is the implicit question of "alignment to what?". AI systems are increasingly used across the world, yet safety alignment is often focused on homogeneous monolingual settings. Additionally, preference training and safety measures often overfit to harms common in Western-centric datasets. Here, we explore the viability of different alignment approaches when balancing dual objectives: addressing and optimizing for a non-homogeneous set of languages and cultural preferences while minimizing both global and local harms. We collect the first set of human annotated red-teaming prompts in different languages distinguishing between global and local harm, which serve as a laboratory for understanding the reliability of alignment techniques when faced with preference distributions that are non-stationary across geographies and languages. While this setting is seldom covered by the literature to date, which primarily centers on English harm mitigation, it captures real-world interactions with AI systems around the world. We establish a new precedent for state-of-the-art alignment techniques across 6 languages with minimal degradation in general performance. Our work provides important insights into cross-lingual transfer and novel optimization approaches to safeguard AI systems designed to serve global populations.
As the development of large language models (LLMs) rapidly advances, securing these models effectively without compromising their utility has become a pivotal area of research. However, current defense strategies against jailbreak attacks (i.e., efforts to bypass security protocols) often suffer from limited adaptability, restricted general capability, and high cost. To address these challenges, we introduce SafeAligner, a methodology implemented at the decoding stage to fortify defenses against jailbreak attacks. We begin by developing two specialized models: the Sentinel Model, which is trained to foster safety, and the Intruder Model, designed to generate riskier responses. SafeAligner leverages the disparity in security levels between the responses from these models to differentiate between harmful and beneficial tokens, effectively guiding the safety alignment by altering the output token distribution of the target model. Extensive experiments show that SafeAligner can increase the likelihood of beneficial tokens, while reducing the occurrence of harmful ones, thereby ensuring secure alignment with minimal loss to generality.
As large language models (LLMs) have advanced rapidly, concerns regarding their safety have become prominent. In this paper, we discover that code-switching in red-teaming queries can effectively elicit undesirable behaviors of LLMs, which are common practices in natural language. We introduce a simple yet effective framework, CSRT, to synthesize code-switching red-teaming queries and investigate the safety and multilingual understanding of LLMs comprehensively. Through extensive experiments with ten state-of-the-art LLMs and code-switching queries combining up to 10 languages, we demonstrate that the CSRT significantly outperforms existing multilingual red-teaming techniques, achieving 46.7% more attacks than standard attacks in English and being effective in conventional safety domains. We also examine the multilingual ability of those LLMs to generate and understand code-switching texts. Additionally, we validate the extensibility of the CSRT by generating code-switching attack prompts with monolingual data. We finally conduct detailed ablation studies exploring code-switching and propound unintended correlation between resource availability of languages and safety alignment in existing multilingual LLMs.
With the widespread application of Large Language Models (LLMs), it has become a significant concern to ensure their safety and prevent harmful responses. While current safe-alignment methods based on instruction fine-tuning and Reinforcement Learning from Human Feedback (RLHF) can effectively reduce harmful responses from LLMs, they often require high-quality datasets and heavy computational overhead during model training. Another way to align language models is to modify the logit of tokens in model outputs without heavy training. Recent studies have shown that contrastive decoding can enhance the performance of language models by reducing the likelihood of confused tokens. However, these methods require the manual selection of contrastive models or instruction templates. To this end, we propose Adversarial Contrastive Decoding (ACD), an optimization-based framework to generate two opposite system prompts for prompt-based contrastive decoding. ACD only needs to apply a lightweight prompt tuning on a rather small anchor dataset (< 3 min for each model) without training the target model. Experiments conducted on extensive models and benchmarks demonstrate that the proposed method achieves much better safety performance than previous model training-free decoding methods without sacrificing its original generation ability.
In this work, we introduce the PKU-SafeRLHF dataset, designed to promote research on safety alignment in large language models (LLMs). As a sibling project to SafeRLHF and BeaverTails, we separate annotations of helpfulness and harmlessness for question-answering pairs, providing distinct perspectives on these coupled attributes. Overall, we provide 44.6k refined prompts and 265k question-answer pairs with safety meta-labels for 19 harm categories and three severity levels ranging from minor to severe, with answers generated by Llama-family models. Based on this, we collected 166.8k preference data, including dual-preference (helpfulness and harmlessness decoupled) and single-preference data (trade-off the helpfulness and harmlessness from scratch), respectively. Using the large-scale annotation data, we further train severity-sensitive moderation for the risk control of LLMs and safety-centric RLHF algorithms for the safety alignment of LLMs. We believe this dataset will be a valuable resource for the community, aiding in the safe deployment of LLMs.
As Artificial General Intelligence (AGI) becomes increasingly integrated into various facets of human life, ensuring the safety and ethical alignment of such systems is paramount. Previous studies primarily focus on single-modality threats, which may not suffice given the integrated and complex nature of cross-modality interactions. We introduce a novel safety alignment challenge called Safe Inputs but Unsafe Output (SIUO) to evaluate cross-modality safety alignment. Specifically, it considers cases where single modalities are safe independently but could potentially lead to unsafe or unethical outputs when combined. To empirically investigate this problem, we developed the SIUO, a cross-modality benchmark encompassing 9 critical safety domains, such as self-harm, illegal activities, and privacy violations. Our findings reveal substantial safety vulnerabilities in both closed- and open-source LVLMs, such as GPT-4V and LLaVA, underscoring the inadequacy of current models to reliably interpret and respond to complex, real-world scenarios.
Large language models (LLMs) are expected to follow instructions from users and engage in conversations. Techniques to enhance LLMs' instruction-following capabilities typically fine-tune them using data structured according to a predefined chat template. Although chat templates are shown to be effective in optimizing LLM performance, their impact on safety alignment of LLMs has been less understood, which is crucial for deploying LLMs safely at scale. In this paper, we investigate how chat templates affect safety alignment of LLMs. We identify a common vulnerability, named ChatBug, that is introduced by chat templates. Our key insight to identify ChatBug is that the chat templates provide a rigid format that need to be followed by LLMs, but not by users. Hence, a malicious user may not necessarily follow the chat template when prompting LLMs. Instead, malicious users could leverage their knowledge of the chat template and accordingly craft their prompts to bypass safety alignments of LLMs. We develop two attacks to exploit the ChatBug vulnerability. We demonstrate that a malicious user can exploit the ChatBug vulnerability of eight state-of-the-art (SOTA) LLMs and effectively elicit unintended responses from these models. Moreover, we show that ChatBug can be exploited by existing jailbreak attacks to enhance their attack success rates. We investigate potential countermeasures to ChatBug. Our results show that while adversarial training effectively mitigates the ChatBug vulnerability, the victim model incurs significant performance degradation. These results highlight the trade-off between safety alignment and helpfulness. Developing new methods for instruction tuning to balance this trade-off is an open and critical direction for future research