Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models

A key component of building safe and reliable language models is enabling the models to appropriately refuse to follow certain instructions or answer certain questions. We may want models to output refusal messages for various categories of user queries, for example, ill-posed questions, instructions for committing illegal acts, or queries which require information past the model's knowledge horizon. Engineering models that refuse to answer such questions is complicated by the fact that an individual may want their model to exhibit varying levels of sensitivity for refusing queries of various categories, and different users may want different refusal rates. The current default approach involves training multiple models with varying proportions of refusal messages from each category to achieve the desired refusal rates, which is computationally expensive and may require training a new model to accommodate each user's desired preference over refusal rates. To address these challenges, we propose refusal tokens, one such token for each refusal category or a single refusal token, which are prepended to the model's responses during training. We then show how to increase or decrease the probability of generating the refusal token for each category during inference to steer the model's refusal behavior. Refusal tokens enable controlling a single model's refusal rates without the need of any further fine-tuning, but only by selectively intervening during generation.

Lottery Ticket Adaptation: Mitigating Destructive Interference in LLMs

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.

Safety Alignment Should Be Made More Than Just a Few Tokens Deep

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.

Teach LLMs to Phish: Stealing Private Information from Language Models

When large language models are trained on private data, it can be a significant privacy risk for them to memorize and regurgitate sensitive information. In this work, we propose a new practical data extraction attack that we call "neural phishing". This attack enables an adversary to target and extract sensitive or personally identifiable information (PII), e.g., credit card numbers, from a model trained on user data with upwards of 10% attack success rates, at times, as high as 50%. Our attack assumes only that an adversary can insert as few as 10s of benign-appearing sentences into the training dataset using only vague priors on the structure of the user data.

Private Fine-tuning of Large Language Models with Zeroth-order Optimization

Fine-tuning large pretrained models on private datasets may run the risk of violating privacy. Differential privacy is a framework for mitigating privacy risks by enforcing algorithmic stability. DP-SGD enables training models with private data in a privacy-preserving manner, but raises new obstacles in the form of performance loss and significant engineering challenges. We introduce DP-ZO, a new method for fine-tuning large language models that preserves the privacy of training data by privatizing zeroth-order optimization. A key insight into the design of our method is that the direction of the gradient in SPSA, the zeroth-order algorithm we use, is always random and the only information that depends on private data is the step size, i.e., a scalar. Therefore, we only need to privatize the scalar step size, which is memory-efficient. DP-ZO, which can be instantiated with either Laplace or Gaussian noise, provides a strong privacy-utility trade-off across different tasks, and model sizes, under conservative privacy budgets. One noteworthy result is that DP-ZO exhibits just $1.86\%$ performance degradation due to privacy at $(1,10^{-5})$-DP when fine-tuning OPT-66B on 1000 training samples from SQuAD.

Visual Adversarial Examples Jailbreak Large Language Models

Recently, there has been a surge of interest in introducing vision into Large Language Models (LLMs). The proliferation of large Visual Language Models (VLMs), such as Flamingo, BLIP-2, and GPT-4, signifies an exciting convergence of advancements in both visual and language foundation models. Yet, the risks associated with this integrative approach are largely unexamined. In this paper, we shed light on the security and safety implications of this trend. First, we underscore that the continuous and high-dimensional nature of the additional visual input space intrinsically makes it a fertile ground for adversarial attacks. This unavoidably expands the attack surfaces of LLMs. Second, we highlight that the broad functionality of LLMs also presents visual attackers with a wider array of achievable adversarial objectives, extending the implications of security failures beyond mere misclassification. To elucidate these risks, we study adversarial examples in the visual input space of a VLM. Specifically, against MiniGPT-4, which incorporates safety mechanisms that can refuse harmful instructions, we present visual adversarial examples that can circumvent the safety mechanisms and provoke harmful behaviors of the model. Remarkably, we discover that adversarial examples, even if optimized on a narrow, manually curated derogatory corpus against specific social groups, can universally jailbreak the model's safety mechanisms. A single such adversarial example can generally undermine MiniGPT-4's safety, enabling it to heed a wide range of harmful instructions and produce harmful content far beyond simply imitating the derogatory corpus used in optimization. Unveiling these risks, we accentuate the urgent need for comprehensive risk assessments, robust defense strategies, and the implementation of responsible practices for the secure and safe utilization of VLMs.

Differentially Private Image Classification by Learning Priors from Random Processes

In privacy-preserving machine learning, differentially private stochastic gradient descent (DP-SGD) performs worse than SGD due to per-sample gradient clipping and noise addition. A recent focus in private learning research is improving the performance of DP-SGD on private data by incorporating priors that are learned on real-world public data. In this work, we explore how we can improve the privacy-utility tradeoff of DP-SGD by learning priors from images generated by random processes and transferring these priors to private data. We propose DP-RandP, a three-phase approach. We attain new state-of-the-art accuracy when training from scratch on CIFAR10, CIFAR100, and MedMNIST for a range of privacy budgets $\varepsilon \in [1, 8]$. In particular, we improve the previous best reported accuracy on CIFAR10 from $60.6 \%$ to $72.3 \%$ for $\varepsilon=1$. Our code is available at https://github.com/inspire-group/DP-RandP.

Differentially Private In-Context Learning

An important question in deploying large language models (LLMs) is how to augment LLMs with private data. We propose Differentially Private In-context Learning (DP-ICL) to enable LLMs to adapt to new tasks while maintaining privacy guarantees. DP-ICL performs private inference by establishing noisy consensus over an ensemble of exemplars using the Report-Noisy-Max mechanism. We evaluate DP-ICL on four benchmarks and find that it achieves comparable performance (<2\% degradation) with non-private ICL.

DP-RAFT: A Differentially Private Recipe for Accelerated Fine-Tuning

A major direction in differentially private machine learning is differentially private fine-tuning: pretraining a model on a source of "public data" and transferring the extracted features to downstream tasks. This is an important setting because many industry deployments fine-tune publicly available feature extractors on proprietary data for downstream tasks. In this paper, we carefully integrate techniques, both new and from prior work, to solve benchmark tasks in computer vision and natural language processing using differentially private fine-tuning. Our key insight is that by accelerating training with the choice of key hyperparameters, we can quickly drive the model parameters to regions in parameter space where the impact of noise is minimized. We obtain new state-of-the art performance on CIFAR10, CIFAR100, FashionMNIST, STL10, and PersonaChat, including $99 \%$ on CIFAR10 for $\varepsilon=1, \delta=1e-5$-DP.

Neurotoxin: Durable Backdoors in Federated Learning

Due to their decentralized nature, federated learning (FL) systems have an inherent vulnerability during their training to adversarial backdoor attacks. In this type of attack, the goal of the attacker is to use poisoned updates to implant so-called backdoors into the learned model such that, at test time, the model's outputs can be fixed to a given target for certain inputs. (As a simple toy example, if a user types "people from New York" into a mobile keyboard app that uses a backdoored next word prediction model, then the model could autocomplete the sentence to "people from New York are rude"). Prior work has shown that backdoors can be inserted into FL models, but these backdoors are often not durable, i.e., they do not remain in the model after the attacker stops uploading poisoned updates. Thus, since training typically continues progressively in production FL systems, an inserted backdoor may not survive until deployment. Here, we propose Neurotoxin, a simple one-line modification to existing backdoor attacks that acts by attacking parameters that are changed less in magnitude during training. We conduct an exhaustive evaluation across ten natural language processing and computer vision tasks, and we find that we can double the durability of state of the art backdoors.