Xiangyu Qi

Xiangyu Qi, Kaixuan Huang, Ashwinee Panda et al.

Recently, there has been a surge of interest in integrating vision into Large Language Models (LLMs), exemplified by Visual Language Models (VLMs) such as Flamingo and GPT-4. This paper sheds light on the security and safety implications of this trend. First, we underscore that the continuous and high-dimensional nature of the visual input makes it a weak link against adversarial attacks, representing an expanded attack surface of vision-integrated LLMs. Second, we highlight that the versatility of LLMs also presents visual attackers with a wider array of achievable adversarial objectives, extending the implications of security failures beyond mere misclassification. As an illustration, we present a case study in which we exploit visual adversarial examples to circumvent the safety guardrail of aligned LLMs with integrated vision. Intriguingly, we discover that a single visual adversarial example can universally jailbreak an aligned LLM, compelling it to heed a wide range of harmful instructions that it otherwise would not) and generate harmful content that transcends the narrow scope of a `few-shot' derogatory corpus initially employed to optimize the adversarial example. Our study underscores the escalating adversarial risks associated with the pursuit of multimodality. Our findings also connect the long-studied adversarial vulnerabilities of neural networks to the nascent field of AI alignment. The presented attack suggests a fundamental adversarial challenge for AI alignment, especially in light of the emerging trend toward multimodality in frontier foundation models.

Tong Wu, Feiran Jia, Xiangyu Qi et al. · princeton

Recently, test-time adaptation (TTA) has been proposed as a promising solution for addressing distribution shifts. It allows a base model to adapt to an unforeseen distribution during inference by leveraging the information from the batch of (unlabeled) test data. However, we uncover a novel security vulnerability of TTA based on the insight that predictions on benign samples can be impacted by malicious samples in the same batch. To exploit this vulnerability, we propose Distribution Invading Attack (DIA), which injects a small fraction of malicious data into the test batch. DIA causes models using TTA to misclassify benign and unperturbed test data, providing an entirely new capability for adversaries that is infeasible in canonical machine learning pipelines. Through comprehensive evaluations, we demonstrate the high effectiveness of our attack on multiple benchmarks across six TTA methods. In response, we investigate two countermeasures to robustify the existing insecure TTA implementations, following the principle of "security by design". Together, we hope our findings can make the community aware of the utility-security tradeoffs in deploying TTA and provide valuable insights for developing robust TTA approaches.

Xiangyu Qi, Tinghao Xie, Jiachen T. Wang et al.

Adversaries can embed backdoors in deep learning models by introducing backdoor poison samples into training datasets. In this work, we investigate how to detect such poison samples to mitigate the threat of backdoor attacks. First, we uncover a post-hoc workflow underlying most prior work, where defenders passively allow the attack to proceed and then leverage the characteristics of the post-attacked model to uncover poison samples. We reveal that this workflow does not fully exploit defenders' capabilities, and defense pipelines built on it are prone to failure or performance degradation in many scenarios. Second, we suggest a paradigm shift by promoting a proactive mindset in which defenders engage proactively with the entire model training and poison detection pipeline, directly enforcing and magnifying distinctive characteristics of the post-attacked model to facilitate poison detection. Based on this, we formulate a unified framework and provide practical insights on designing detection pipelines that are more robust and generalizable. Third, we introduce the technique of Confusion Training (CT) as a concrete instantiation of our framework. CT applies an additional poisoning attack to the already poisoned dataset, actively decoupling benign correlation while exposing backdoor patterns to detection. Empirical evaluations on 4 datasets and 14 types of attacks validate the superiority of CT over 14 baseline defenses.

Xiangyu Qi, Yi Zeng, Tinghao Xie et al.

Optimizing large language models (LLMs) for downstream use cases often involves the customization of pre-trained LLMs through further fine-tuning. Meta's open release of Llama models and OpenAI's APIs for fine-tuning GPT-3.5 Turbo on custom datasets also encourage this practice. But, what are the safety costs associated with such custom fine-tuning? We note that while existing safety alignment infrastructures can restrict harmful behaviors of LLMs at inference time, they do not cover safety risks when fine-tuning privileges are extended to end-users. Our red teaming studies find that the safety alignment of LLMs can be compromised by fine-tuning with only a few adversarially designed training examples. For instance, we jailbreak GPT-3.5 Turbo's safety guardrails by fine-tuning it on only 10 such examples at a cost of less than $0.20 via OpenAI's APIs, making the model responsive to nearly any harmful instructions. Disconcertingly, our research also reveals that, even without malicious intent, simply fine-tuning with benign and commonly used datasets can also inadvertently degrade the safety alignment of LLMs, though to a lesser extent. These findings suggest that fine-tuning aligned LLMs introduces new safety risks that current safety infrastructures fall short of addressing -- even if a model's initial safety alignment is impeccable, it is not necessarily to be maintained after custom fine-tuning. We outline and critically analyze potential mitigations and advocate for further research efforts toward reinforcing safety protocols for the custom fine-tuning of aligned LLMs.

Tinghao Xie, Xiangyu Qi, Ping He et al.

We present a novel defense, against backdoor attacks on Deep Neural Networks (DNNs), wherein adversaries covertly implant malicious behaviors (backdoors) into DNNs. Our defense falls within the category of post-development defenses that operate independently of how the model was generated. The proposed defense is built upon a novel reverse engineering approach that can directly extract backdoor functionality of a given backdoored model to a backdoor expert model. The approach is straightforward -- finetuning the backdoored model over a small set of intentionally mislabeled clean samples, such that it unlearns the normal functionality while still preserving the backdoor functionality, and thus resulting in a model (dubbed a backdoor expert model) that can only recognize backdoor inputs. Based on the extracted backdoor expert model, we show the feasibility of devising highly accurate backdoor input detectors that filter out the backdoor inputs during model inference. Further augmented by an ensemble strategy with a finetuned auxiliary model, our defense, BaDExpert (Backdoor Input Detection with Backdoor Expert), effectively mitigates 17 SOTA backdoor attacks while minimally impacting clean utility. The effectiveness of BaDExpert has been verified on multiple datasets (CIFAR10, GTSRB and ImageNet) across various model architectures (ResNet, VGG, MobileNetV2 and Vision Transformer).

Xiangyu Qi, Tinghao Xie, Yiming Li et al.

Recent studies revealed that deep learning is susceptible to backdoor poisoning attacks. An adversary can embed a hidden backdoor into a model to manipulate its predictions by only modifying a few training data, without controlling the training process. Currently, a tangible signature has been widely observed across a diverse set of backdoor poisoning attacks -- models trained on a poisoned dataset tend to learn separable latent representations for poison and clean samples. This latent separation is so pervasive that a family of backdoor defenses directly take it as a default assumption (dubbed latent separability assumption), based on which to identify poison samples via cluster analysis in the latent space. An intriguing question consequently follows: is the latent separation unavoidable for backdoor poisoning attacks? This question is central to understanding whether the assumption of latent separability provides a reliable foundation for defending against backdoor poisoning attacks. In this paper, we design adaptive backdoor poisoning attacks to present counter-examples against this assumption. Our methods include two key components: (1) a set of trigger-planted samples correctly labeled to their semantic classes (other than the target class) that can regularize backdoor learning; (2) asymmetric trigger planting strategies that help to boost attack success rate (ASR) as well as to diversify latent representations of poison samples. Extensive experiments on benchmark datasets verify the effectiveness of our adaptive attacks in bypassing existing latent separation based backdoor defenses. Moreover, our attacks still maintain a high attack success rate with negligible clean accuracy drop. Our studies call for defense designers to take caution when leveraging latent separation as an assumption in their defenses.

Boyi Wei, Kaixuan Huang, Yangsibo Huang et al. · princeton

Large language models (LLMs) show inherent brittleness in their safety mechanisms, as evidenced by their susceptibility to jailbreaking and even non-malicious fine-tuning. This study explores this brittleness of safety alignment by leveraging pruning and low-rank modifications. We develop methods to identify critical regions that are vital for safety guardrails, and that are disentangled from utility-relevant regions at both the neuron and rank levels. Surprisingly, the isolated regions we find are sparse, comprising about $3\%$ at the parameter level and $2.5\%$ at the rank level. Removing these regions compromises safety without significantly impacting utility, corroborating the inherent brittleness of the model's safety mechanisms. Moreover, we show that LLMs remain vulnerable to low-cost fine-tuning attacks even when modifications to the safety-critical regions are restricted. These findings underscore the urgent need for more robust safety strategies in LLMs.

Ashwinee Panda, Berivan Isik, Xiangyu Qi et al.

Existing methods for adapting large language models (LLMs) to new tasks are not suited to multi-task adaptation because they modify all the model weights -- causing destructive interference between tasks. The resulting effects, such as catastrophic forgetting of earlier tasks, make it challenging to obtain good performance on multiple tasks at the same time. To mitigate this, we propose Lottery Ticket Adaptation (LoTA), a sparse adaptation method that identifies and optimizes only a sparse subnetwork of the model. We evaluate LoTA on a wide range of challenging tasks such as instruction following, reasoning, math, and summarization. LoTA obtains better performance than full fine-tuning and low-rank adaptation (LoRA), and maintains good performance even after training on other tasks -- thus, avoiding catastrophic forgetting. By extracting and fine-tuning over lottery tickets (or sparse task vectors), LoTA also enables model merging over highly dissimilar tasks. Our code is made publicly available at https://github.com/kiddyboots216/lottery-ticket-adaptation.

Xiangyu Qi, Boyi Wei, Nicholas Carlini et al. · princeton

Stakeholders -- from model developers to policymakers -- seek to minimize the dual-use risks of large language models (LLMs). An open challenge to this goal is whether technical safeguards can impede the misuse of LLMs, even when models are customizable via fine-tuning or when model weights are fully open. In response, several recent studies have proposed methods to produce durable LLM safeguards for open-weight LLMs that can withstand adversarial modifications of the model's weights via fine-tuning. This holds the promise of raising adversaries' costs even under strong threat models where adversaries can directly fine-tune model weights. However, in this paper, we urge for more careful characterization of the limits of these approaches. Through several case studies, we demonstrate that even evaluating these defenses is exceedingly difficult and can easily mislead audiences into thinking that safeguards are more durable than they really are. We draw lessons from the evaluation pitfalls that we identify and suggest future research carefully cabin claims to more constrained, well-defined, and rigorously examined threat models, which can provide more useful and candid assessments to stakeholders.

Jiongxiao Wang, Jiazhao Li, Yiquan Li et al.

Despite the general capabilities of Large Language Models (LLM), these models still request fine-tuning or adaptation with customized data when meeting specific business demands. However, this process inevitably introduces new threats, particularly against the Fine-tuning based Jailbreak Attack (FJAttack) under the setting of Language-Model-as-a-Service (LMaaS), where the model's safety has been significantly compromised by fine-tuning users' uploaded examples contain just a few harmful examples. Though potential defenses have been proposed that the service providers can integrate safety examples into the fine-tuning dataset to reduce safety issues, such approaches require incorporating a substantial amount of data, making it inefficient. To effectively defend against the FJAttack with limited safety examples under LMaaS, we propose the Backdoor Enhanced Safety Alignment method inspired by an analogy with the concept of backdoor attacks. In particular, service providers will construct prefixed safety examples with a secret prompt, acting as a "backdoor trigger". By integrating prefixed safety examples into the fine-tuning dataset, the subsequent fine-tuning process effectively acts as the "backdoor attack", establishing a strong correlation between the secret prompt and safety generations. Consequently, safe responses are ensured once service providers prepend this secret prompt ahead of any user input during inference. Our comprehensive experiments demonstrate that through the Backdoor Enhanced Safety Alignment with adding as few as 11 prefixed safety examples, the maliciously fine-tuned LLMs will achieve similar safety performance as the original aligned models without harming the benign performance. Furthermore, we also present the effectiveness of our method in a more practical setting where the fine-tuning data consists of both FJAttack examples and the fine-tuning task data.

Haonan Li, Xudong Han, Zenan Zhai et al.

To address this gap, we introduce Libra-Leaderboard, a comprehensive framework designed to rank LLMs through a balanced evaluation of performance and safety. Combining a dynamic leaderboard with an interactive LLM arena, Libra-Leaderboard encourages the joint optimization of capability and safety. Unlike traditional approaches that average performance and safety metrics, Libra-Leaderboard uses a distance-to-optimal-score method to calculate the overall rankings. This approach incentivizes models to achieve a balance rather than excelling in one dimension at the expense of some other ones. In the first release, Libra-Leaderboard evaluates 26 mainstream LLMs from 14 leading organizations, identifying critical safety challenges even in state-of-the-art models.

Tinghao Xie, Xiangyu Qi, Yi Zeng et al.

Evaluating aligned large language models' (LLMs) ability to recognize and reject unsafe user requests is crucial for safe, policy-compliant deployments. Existing evaluation efforts, however, face three limitations that we address with SORRY-Bench, our proposed benchmark. First, existing methods often use coarse-grained taxonomies of unsafe topics, and are over-representing some fine-grained topics. For example, among the ten existing datasets that we evaluated, tests for refusals of self-harm instructions are over 3x less represented than tests for fraudulent activities. SORRY-Bench improves on this by using a fine-grained taxonomy of 44 potentially unsafe topics, and 440 class-balanced unsafe instructions, compiled through human-in-the-loop methods. Second, linguistic characteristics and formatting of prompts are often overlooked, like different languages, dialects, and more -- which are only implicitly considered in many evaluations. We supplement SORRY-Bench with 20 diverse linguistic augmentations to systematically examine these effects. Third, existing evaluations rely on large LLMs (e.g., GPT-4) for evaluation, which can be computationally expensive. We investigate design choices for creating a fast, accurate automated safety evaluator. By collecting 7K+ human annotations and conducting a meta-evaluation of diverse LLM-as-a-judge designs, we show that fine-tuned 7B LLMs can achieve accuracy comparable to GPT-4 scale LLMs, with lower computational cost. Putting these together, we evaluate over 50 proprietary and open-weight LLMs on SORRY-Bench, analyzing their distinctive safety refusal behaviors. We hope our effort provides a building block for systematic evaluations of LLMs' safety refusal capabilities, in a balanced, granular, and efficient manner. Benchmark demo, data, code, and models are available through https://sorry-bench.github.io.

Xiangyu Qi, Ashwinee Panda, Kaifeng Lyu et al.

The safety alignment of current Large Language Models (LLMs) is vulnerable. Relatively simple attacks, or even benign fine-tuning, can jailbreak aligned models. We argue that many of these vulnerabilities are related to a shared underlying issue: safety alignment can take shortcuts, wherein the alignment adapts a model's generative distribution primarily over only its very first few output tokens. We refer to this issue as shallow safety alignment. In this paper, we present case studies to explain why shallow safety alignment can exist and provide evidence that current aligned LLMs are subject to this issue. We also show how these findings help explain multiple recently discovered vulnerabilities in LLMs, including the susceptibility to adversarial suffix attacks, prefilling attacks, decoding parameter attacks, and fine-tuning attacks. Importantly, we discuss how this consolidated notion of shallow safety alignment sheds light on promising research directions for mitigating these vulnerabilities. For instance, we show that deepening the safety alignment beyond just the first few tokens can often meaningfully improve robustness against some common exploits. Finally, we design a regularized finetuning objective that makes the safety alignment more persistent against fine-tuning attacks by constraining updates on initial tokens. Overall, we advocate that future safety alignment should be made more than just a few tokens deep.