The current paradigm for safety alignment of large language models (LLMs) follows a one-size-fits-all approach: the model refuses to interact with any content deemed unsafe by the model provider. This approach lacks flexibility in the face of varying social norms across cultures and regions. In addition, users may have diverse safety needs, making a model with static safety standards too restrictive to be useful, as well as too costly to be re-aligned. We propose Controllable Safety Alignment (CoSA), a framework designed to adapt models to diverse safety requirements without re-training. Instead of aligning a fixed model, we align models to follow safety configs -- free-form natural language descriptions of the desired safety behaviors -- that are provided as part of the system prompt. To adjust model safety behavior, authorized users only need to modify such safety configs at inference time. To enable that, we propose CoSAlign, a data-centric method for aligning LLMs to easily adapt to diverse safety configs. Furthermore, we devise a novel controllability evaluation protocol that considers both helpfulness and configured safety, summarizing them into CoSA-Score, and construct CoSApien, a human-authored benchmark that consists of real-world LLM use cases with diverse safety requirements and corresponding evaluation prompts. We show that CoSAlign leads to substantial gains of controllability over strong baselines including in-context alignment. Our framework encourages better representation and adaptation to pluralistic human values in LLMs, and thereby increasing their practicality.
How to defend large language models (LLMs) from generating toxic content is an important research area. Yet, most research focused on various model training techniques to remediate LLMs by updating their weights. A typical related research area is safety alignment. This however is often costly and tedious and can expose the model to even more problems such as catastrophic forgetting if the trainings are not carefully handled by experienced NLP practitioners. We thus propose a simple yet effective and novel algorithm, namely \textbf{Tox}ic Subword \textbf{Prun}ing (ToxPrune) to prune the subword contained by the toxic words from BPE in trained LLMs. In contrast to the previous work that demonstrates pruning BPE tokens as harmful to the task of machine translation, we surprisingly found its usefulness in preventing toxic content from being generated on LLMs. Fortunately, our findings suggest that ToxPrune simultaneously improves the toxic language model NSFW-3B on the task of dialogue response generation obviously. We surprisingly found that ToxPrune can even obviously improve official Llama-3.1-6B in the metric of dialogue diversity. Extensive automatic results and human evaluation indicate that ToxPrune could be helpful for both remediating toxic LLMs and improving non-toxic LLMs on the task of dialogue response generation.\footnote{We plan to release the resources to facilitate future work.}
Data is a crucial element in large language model (LLM) alignment. Recent studies have explored using LLMs for efficient data collection. However, LLM-generated data often suffers from quality issues, with underrepresented or absent aspects and low-quality datapoints. To address these problems, we propose Data Advisor, an enhanced LLM-based method for generating data that takes into account the characteristics of the desired dataset. Starting from a set of pre-defined principles in hand, Data Advisor monitors the status of the generated data, identifies weaknesses in the current dataset, and advises the next iteration of data generation accordingly. Data Advisor can be easily integrated into existing data generation methods to enhance data quality and coverage. Experiments on safety alignment of three representative LLMs (i.e., Mistral, Llama2, and Falcon) demonstrate the effectiveness of Data Advisor in enhancing model safety against various fine-grained safety issues without sacrificing model utility.
Large language models (LLMs) have a transformative impact on a variety of scientific tasks across disciplines including biology, chemistry, medicine, and physics. However, ensuring the safety alignment of these models in scientific research remains an underexplored area, with existing benchmarks primarily focusing on textual content and overlooking key scientific representations such as molecular, protein, and genomic languages. Moreover, the safety mechanisms of LLMs in scientific tasks are insufficiently studied. To address these limitations, we introduce SciSafeEval, a comprehensive benchmark designed to evaluate the safety alignment of LLMs across a range of scientific tasks. SciSafeEval spans multiple scientific languages-including textual, molecular, protein, and genomic-and covers a wide range of scientific domains. We evaluate LLMs in zero-shot, few-shot and chain-of-thought settings, and introduce a "jailbreak" enhancement feature that challenges LLMs equipped with safety guardrails, rigorously testing their defenses against malicious intention. Our benchmark surpasses existing safety datasets in both scale and scope, providing a robust platform for assessing the safety and performance of LLMs in scientific contexts. This work aims to facilitate the responsible development and deployment of LLMs, promoting alignment with safety and ethical standards in scientific research.
Safety is a paramount concern for large language models (LLMs) in open deployment, motivating the development of safeguard methods that enforce ethical and responsible use through safety alignment or guardrail mechanisms. Jailbreak attacks that exploit the \emph{false negatives} of safeguard methods have emerged as a prominent research focus in the field of LLM security. However, we found that the malicious attackers could also exploit false positives of safeguards, i.e., fooling the safeguard model to block safe content mistakenly, leading to a denial-of-service (DoS) affecting LLM users. To bridge the knowledge gap of this overlooked threat, we explore multiple attack methods that include inserting a short adversarial prompt into user prompt templates and corrupting the LLM on the server by poisoned fine-tuning. In both ways, the attack triggers safeguard rejections of user requests from the client. Our evaluation demonstrates the severity of this threat across multiple scenarios. For instance, in the scenario of white-box adversarial prompt injection, the attacker can use our optimization process to automatically generate seemingly safe adversarial prompts, approximately only 30 characters long, that universally block over 97% of user requests on Llama Guard 3. These findings reveal a new dimension in LLM safeguard evaluation -- adversarial robustness to false positives.
While large language models (LLMs) have seen unprecedented advancements in capabilities and applications across a variety of use-cases, safety alignment of these models is still an area of active research. The fragile nature of LLMs, even models that have undergone extensive alignment and safety training regimes, warrants additional safety steering steps via training-free, inference-time methods. While recent work in the area of mechanistic interpretability has investigated how activations in latent representation spaces may encode concepts, and thereafter performed representation engineering to induce such concepts in LLM outputs, the applicability of such for safety is relatively under-explored. Unlike recent inference-time safety steering works, in this paper we explore safety steering of LLM outputs using: (i) category-specific steering vectors, thereby enabling fine-grained control over the steering, and (ii) sophisticated methods for extracting informative steering vectors for more effective safety steering while retaining quality of the generated text. We demonstrate our exploration on multiple LLMs and datasets, and showcase the effectiveness of the proposed steering method, along with a discussion on the implications and best practices.
Despite significant ongoing efforts in safety alignment, large language models (LLMs) such as GPT-4 and LLaMA 3 remain vulnerable to jailbreak attacks that can induce harmful behaviors, including through the use of adversarial suffixes. Building on prior research, we hypothesize that these adversarial suffixes are not mere bugs but may represent features that can dominate the LLM's behavior. To evaluate this hypothesis, we conduct several experiments. First, we demonstrate that benign features can be effectively made to function as adversarial suffixes, i.e., we develop a feature extraction method to extract sample-agnostic features from benign dataset in the form of suffixes and show that these suffixes may effectively compromise safety alignment. Second, we show that adversarial suffixes generated from jailbreak attacks may contain meaningful features, i.e., appending the same suffix to different prompts results in responses exhibiting specific characteristics. Third, we show that such benign-yet-safety-compromising features can be easily introduced through fine-tuning using only benign datasets. As a result, we are able to completely eliminate GPT's safety alignment in a blackbox setting through finetuning with only benign data. Our code and data is available at \url{https://github.com/suffix-maybe-feature/adver-suffix-maybe-features}.
LLMs(Large Language Models) nowadays have widespread adoption as a tool for solving issues across various domain/tasks. These models since are susceptible to produce harmful or toxic results, inference-time adversarial attacks, therefore they do undergo safety alignment training and Red teaming for putting in safety guardrails. For using these models, usually fine-tuning is done for model alignment on the desired tasks, which can make model more aligned but also make it more susceptible to produce unsafe responses, if fine-tuned with harmful data.In this paper, we study how much of impact introduction of harmful data in fine-tuning can make, and if it can override the safety protection of those models. Conversely,it was also explored that if model is fine-tuned on safety data can make the model produce more safer responses. Further we explore if fine-tuning the model on harmful data makes it less helpful or less trustworthy because of increase in model uncertainty leading to knowledge drift. Our extensive experimental results shown that Safety protection in an open-source can be overridden, when fine-tuned with harmful data as observed by ASR increasing by 35% when compared to basemodel's ASR. Also, as observed, fine-tuning a model with harmful data made the harmful fine-tuned model highly uncertain with huge knowledge drift and less truthfulness in its responses. Furthermore, for the safe fine-tuned model, ASR decreases by 51.68% as compared to the basemodel, and Safe model also shown in minor drop in uncertainty and truthfulness as compared to basemodel. This paper's code is available at: https://github.com/techsachinkr/Overriding_Model_Safety_Protections
Recent research demonstrates that the nascent fine-tuning-as-a-service business model exposes serious safety concerns -- fine-tuning over a few harmful data uploaded by the users can compromise the safety alignment of the model. The attack, known as harmful fine-tuning attack, has raised a broad research interest among the community. However, as the attack is still new, \textbf{we observe that there are general misunderstandings within the research community.} To clear up concern, this paper provide a comprehensive overview to three aspects of harmful fine-tuning: attacks setting, defense design and evaluation methodology. Specifically, we first present the threat model of the problem, and introduce the harmful fine-tuning attack and its variants. Then we systematically survey the existing literature on attacks/defenses/mechanical analysis of the problem. Finally, we introduce the evaluation methodology and outline future research directions that might contribute to the development of the field. Additionally, we present a list of questions of interest, which might be useful to refer to when reviewers in the peer review process question the realism of the experiment/attack/defense setting. A curated list of relevant papers is maintained and made accessible at: https://github.com/git-disl/awesome_LLM-harmful-fine-tuning-papers.
In recent years, Large Language Models (LLMs) have gained widespread use, raising concerns about their security. Traditional jailbreak attacks, which often rely on the model internal information or have limitations when exploring the unsafe behavior of the victim model, limiting their reducing their general applicability. In this paper, we introduce PathSeeker, a novel black-box jailbreak method, which is inspired by the game of rats escaping a maze. We think that each LLM has its unique "security maze", and attackers attempt to find the exit learning from the received feedback and their accumulated experience to compromise the target LLM's security defences. Our approach leverages multi-agent reinforcement learning, where smaller models collaborate to guide the main LLM in performing mutation operations to achieve the attack objectives. By progressively modifying inputs based on the model's feedback, our system induces richer, harmful responses. During our manual attempts to perform jailbreak attacks, we found that the vocabulary of the response of the target model gradually became richer and eventually produced harmful responses. Based on the observation, we also introduce a reward mechanism that exploits the expansion of vocabulary richness in LLM responses to weaken security constraints. Our method outperforms five state-of-the-art attack techniques when tested across 13 commercial and open-source LLMs, achieving high attack success rates, especially in strongly aligned commercial models like GPT-4o-mini, Claude-3.5, and GLM-4-air with strong safety alignment. This study aims to improve the understanding of LLM security vulnerabilities and we hope that this sturdy can contribute to the development of more robust defenses.
Text generation has a fundamental limitation almost by definition: there is no taking back tokens that have been generated, even when they are clearly problematic. In the context of language model safety, when a partial unsafe generation is produced, language models by their nature tend to happily keep on generating similarly unsafe additional text. This is in fact how safety alignment of frontier models gets circumvented in the wild, despite great efforts in improving their safety. Deviating from the paradigm of approaching safety alignment as prevention (decreasing the probability of harmful responses), we propose backtracking, a technique that allows language models to "undo" and recover from their own unsafe generation through the introduction of a special [RESET] token. Our method can be incorporated into either SFT or DPO training to optimize helpfulness and harmlessness. We show that models trained to backtrack are consistently safer than baseline models: backtracking Llama-3-8B is four times more safe than the baseline model (6.1\% $\to$ 1.5\%) in our evaluations without regression in helpfulness. Our method additionally provides protection against four adversarial attacks including an adaptive attack, despite not being trained to do so.
While there has been progress towards aligning Large Language Models (LLMs) with human values and ensuring safe behaviour at inference time, safety-guards can easily be removed when fine-tuned on unsafe and harmful datasets.While this setting has been treated extensively, another popular training paradigm, learning from unsafe feedback with reinforcement learning, has previously been unexplored. This is concerning due to the widespread deployment of feedback collection systems. We address this gap by providing an analysis of learning settings where feedback is adversarial and noisy, i.e. that unsafe samples are preferred over safe ones despite model developers goal to maintain safety. We find that safety-aligned LLMs easily explore unsafe action spaces through generating harmful text and optimize for adversarial reward indicating that current safety guards are not enough to prevent learning from unsafe feedback. In order to protect against this vulnerability, we adapt a number of both "implict" and "explicit" harmful fine-tuning defences to evaluate whether they are effective as learning constraints in an RL setting finding that no method is generally effective pointing to the need for more research in defences given the widespread adoption of methods designed to learn from feedback. We end the paper with the observation that some defences work by performing "harmless reward hacking" for which we provide a theoretical explanation drawn from the theory of Constrained Markov Decision Processes and provide some direction for future defence development.
We introduce Synthetic Alignment data Generation for Safety Evaluation and Red Teaming (SAGE-RT or SAGE) a novel pipeline for generating synthetic alignment and red-teaming data. Existing methods fall short in creating nuanced and diverse datasets, providing necessary control over the data generation and validation processes, or require large amount of manually generated seed data. SAGE addresses these limitations by using a detailed taxonomy to produce safety-alignment and red-teaming data across a wide range of topics. We generated 51,000 diverse and in-depth prompt-response pairs, encompassing over 1,500 topics of harmfulness and covering variations of the most frequent types of jailbreaking prompts faced by large language models (LLMs). We show that the red-teaming data generated through SAGE jailbreaks state-of-the-art LLMs in more than 27 out of 32 sub-categories, and in more than 58 out of 279 leaf-categories (sub-sub categories). The attack success rate for GPT-4o, GPT-3.5-turbo is 100% over the sub-categories of harmfulness. Our approach avoids the pitfalls of synthetic safety-training data generation such as mode collapse and lack of nuance in the generation pipeline by ensuring a detailed coverage of harmful topics using iterative expansion of the topics and conditioning the outputs on the generated raw-text. This method can be used to generate red-teaming and alignment data for LLM Safety completely synthetically to make LLMs safer or for red-teaming the models over a diverse range of topics.
Safety alignment is indispensable for Large Language Models (LLMs) to defend threats from malicious instructions. However, recent researches reveal safety-aligned LLMs prone to reject benign queries due to the exaggerated safety issue, limiting their helpfulness. In this paper, we propose a Safety-Conscious Activation Steering (SCANS) method to mitigate the exaggerated safety concerns in aligned LLMs. First, SCANS extracts the refusal steering vectors within the activation space and utilizes vocabulary projection to anchor some specific safety-critical layers which influence model refusal behavior. Second, by tracking the hidden state transition, SCANS identifies the steering direction and steers the model behavior accordingly, achieving a balance between exaggerated safety and adequate safety. Experiments show that SCANS achieves new state-of-the-art performance on XSTest and OKTest benchmarks, without impairing their defense capability against harmful queries and maintaining almost unchanged model capability.
Despite prior safety alignment efforts, mainstream LLMs can still generate harmful and unethical content when subjected to jailbreaking attacks. Existing jailbreaking methods fall into two main categories: template-based and optimization-based methods. The former requires significant manual effort and domain knowledge, while the latter, exemplified by Greedy Coordinate Gradient (GCG), which seeks to maximize the likelihood of harmful LLM outputs through token-level optimization, also encounters several limitations: requiring white-box access, necessitating pre-constructed affirmative phrase, and suffering from low efficiency. In this paper, we present ECLIPSE, a novel and efficient black-box jailbreaking method utilizing optimizable suffixes. Drawing inspiration from LLMs' powerful generation and optimization capabilities, we employ task prompts to translate jailbreaking goals into natural language instructions. This guides the LLM to generate adversarial suffixes for malicious queries. In particular, a harmfulness scorer provides continuous feedback, enabling LLM self-reflection and iterative optimization to autonomously and efficiently produce effective suffixes. Experimental results demonstrate that ECLIPSE achieves an average attack success rate (ASR) of 0.92 across three open-source LLMs and GPT-3.5-Turbo, significantly surpassing GCG in 2.4 times. Moreover, ECLIPSE is on par with template-based methods in ASR while offering superior attack efficiency, reducing the average attack overhead by 83%.
In today's era, where large language models (LLMs) are integrated into numerous real-world applications, ensuring their safety and robustness is crucial for responsible AI usage. Automated red-teaming methods play a key role in this process by generating adversarial attacks to identify and mitigate potential vulnerabilities in these models. However, existing methods often struggle with slow performance, limited categorical diversity, and high resource demands. While Rainbow Teaming, a recent approach, addresses the diversity challenge by framing adversarial prompt generation as a quality-diversity search, it remains slow and requires a large fine-tuned mutator for optimal performance. To overcome these limitations, we propose Ferret, a novel approach that builds upon Rainbow Teaming by generating multiple adversarial prompt mutations per iteration and using a scoring function to rank and select the most effective adversarial prompt. We explore various scoring functions, including reward models, Llama Guard, and LLM-as-a-judge, to rank adversarial mutations based on their potential harm to improve the efficiency of the search for harmful mutations. Our results demonstrate that Ferret, utilizing a reward model as a scoring function, improves the overall attack success rate (ASR) to 95%, which is 46% higher than Rainbow Teaming. Additionally, Ferret reduces the time needed to achieve a 90% ASR by 15.2% compared to the baseline and generates adversarial prompts that are transferable i.e. effective on other LLMs of larger size. Our codes are available at https://github.com/declare-lab/ferret.
Large Language Models (LLMs) are implicit troublemakers. While they provide valuable insights and assist in problem-solving, they can also potentially serve as a resource for malicious activities. Implementing safety alignment could mitigate the risk of LLMs generating harmful responses. We argue that: even when an LLM appears to successfully block harmful queries, there may still be hidden vulnerabilities that could act as ticking time bombs. To identify these underlying weaknesses, we propose to use a cost value model as both a detector and an attacker. Trained on external or self-generated harmful datasets, the cost value model could successfully influence the original safe LLM to output toxic content in decoding process. For instance, LLaMA-2-chat 7B outputs 39.18% concrete toxic content, along with only 22.16% refusals without any harmful suffixes. These potential weaknesses can then be exploited via prompt optimization such as soft prompts on images. We name this decoding strategy: Jailbreak Value Decoding (JVD), emphasizing that seemingly secure LLMs may not be as safe as we initially believe. They could be used to gather harmful data or launch covert attacks.
Safety aligned Large Language Models (LLMs) are vulnerable to harmful fine-tuning attacks \cite{qi2023fine}-- a few harmful data mixed in the fine-tuning dataset can break the LLMs's safety alignment. Existing mitigation strategies include alignment stage solutions \cite{huang2024vaccine, rosati2024representation} and fine-tuning stage solutions \cite{huang2024lazy,mukhoti2023fine}. However, our evaluation shows that both categories of defenses fail \textit{when some specific training hyper-parameters are chosen} -- a large learning rate or a large number of training epochs in the fine-tuning stage can easily invalidate the defense, which however, is necessary to guarantee finetune performance. To this end, we propose Antidote, a post-fine-tuning stage solution, which remains \textbf{\textit{agnostic to the training hyper-parameters in the fine-tuning stage}}. Antidote relies on the philosophy that by removing the harmful parameters, the harmful model can be recovered from the harmful behaviors, regardless of how those harmful parameters are formed in the fine-tuning stage. With this philosophy, we introduce a one-shot pruning stage after harmful fine-tuning to remove the harmful weights that are responsible for the generation of harmful content. Despite its embarrassing simplicity, empirical results show that Antidote can reduce harmful score while maintaining accuracy on downstream tasks.Our project page is at \url{https://huangtiansheng.github.io/Antidote_gh_page/}
Large language models are susceptible to jailbreak attacks, which can result in the generation of harmful content. While prior defenses mitigate these risks by perturbing or inspecting inputs, they ignore competing objectives, the underlying cause of alignment failures. In this paper, we propose Alignment-Enhanced Decoding (AED), a novel defense that employs adaptive decoding to address the root causes of jailbreak issues. We first define the Competitive Index to quantify alignment failures and utilize feedback from self-evaluation to compute post-alignment logits. Then, AED adaptively combines AED and post-alignment logits with the original logits to obtain harmless and helpful distributions. Consequently, our method enhances safety alignment while maintaining helpfulness. We conduct experiments across five models and four common jailbreaks, with the results validating the effectiveness of our approach. Code is available at https://github.com/GIGABaozi/AED.git.
As Large Language Models (LLMs) are increasingly being deployed in safety-critical applications, their vulnerability to potential jailbreaks -- malicious prompts that can disable the safety mechanism of LLMs -- has attracted growing research attention. While alignment methods have been proposed to protect LLMs from jailbreaks, many have found that aligned LLMs can still be jailbroken by carefully crafted malicious prompts, producing content that violates policy regulations. Existing jailbreak attacks on LLMs can be categorized into prompt-level methods which make up stories/logic to circumvent safety alignment and token-level attack methods which leverage gradient methods to find adversarial tokens. In this work, we introduce the concept of Ensemble Jailbreak and explore methods that can integrate prompt-level and token-level jailbreak into a more powerful hybrid jailbreak attack. Specifically, we propose a novel EnJa attack to hide harmful instructions using prompt-level jailbreak, boost the attack success rate using a gradient-based attack, and connect the two types of jailbreak attacks via a template-based connector. We evaluate the effectiveness of EnJa on several aligned models and show that it achieves a state-of-the-art attack success rate with fewer queries and is much stronger than any individual jailbreak.