Eric Pauley

Kyle Domico, Jean-Charles Noirot Ferrand, Ryan Sheatsley et al.

Attacks on machine learning models have been extensively studied through stateless optimization. In this paper, we demonstrate how a reinforcement learning (RL) agent can learn a new class of attack algorithms that generate adversarial samples. Unlike traditional adversarial machine learning (AML) methods that craft adversarial samples independently, our RL-based approach retains and exploits past attack experience to improve the effectiveness and efficiency of future attacks. We formulate adversarial sample generation as a Markov Decision Process and evaluate RL's ability to (a) learn effective and efficient attack strategies and (b) compete with state-of-the-art AML. On two image classification benchmarks, our agent increases attack success rate by up to 13.2% and decreases the average number of victim model queries per attack by up to 16.9% from the start to the end of training. In a head-to-head comparison with state-of-the-art image attacks, our approach enables an adversary to generate adversarial samples with 17% more success on unseen inputs post-training. From a security perspective, this work demonstrates a powerful new attack vector that uses RL to train agents that attack ML models efficiently and at scale.

Kunyang Li, Jean-Charles Noirot Ferrand, Ryan Sheatsley et al.

Fine-tuning has become the standard practice for adapting pre-trained models to downstream tasks. However, the impact on model robustness is not well understood. In this work, we characterize the robustness-accuracy trade-off in fine-tuning. We evaluate the robustness and accuracy of fine-tuned models over 6 benchmark datasets and 7 different fine-tuning strategies. We observe a consistent trade-off between adversarial robustness and accuracy. Peripheral updates such as BitFit are more effective for simple tasks -- over 75% above the average measured by the area under the Pareto frontiers on CIFAR-10 and CIFAR-100. In contrast, fine-tuning information-heavy layers, such as attention layers via Compacter, achieves a better Pareto frontier on more complex tasks -- 57.5% and 34.6% above the average on Caltech-256 and CUB-200, respectively. Lastly, we observe that the robustness of fine-tuning against out-of-distribution data closely tracks accuracy. These insights emphasize the need for robustness-aware fine-tuning to ensure reliable real-world deployments.

Jean-Charles Noirot Ferrand, Yohan Beugin, Eric Pauley et al.

Alignment in large language models (LLMs) is used to enforce guidelines such as safety. Yet, alignment fails in the face of jailbreak attacks that modify inputs to induce unsafe outputs. In this paper, we introduce and evaluate a new technique for jailbreak attacks. We observe that alignment embeds a safety classifier in the LLM responsible for deciding between refusal and compliance, and seek to extract an approximation of this classifier: a surrogate classifier. To this end, we build candidate classifiers from subsets of the LLM. We first evaluate the degree to which candidate classifiers approximate the LLM's safety classifier in benign and adversarial settings. Then, we attack the candidates and measure how well the resulting adversarial inputs transfer to the LLM. Our evaluation shows that the best candidates achieve accurate agreement (an F1 score above 80%) using as little as 20% of the model architecture. Further, we find that attacks mounted on the surrogate classifiers can be transferred to the LLM with high success. For example, a surrogate using only 50% of the Llama 2 model achieved an attack success rate (ASR) of 70% with half the memory footprint and runtime -- a substantial improvement over attacking the LLM directly, where we only observed a 22% ASR. These results show that extracting surrogate classifiers is an effective and efficient means for modeling (and therein addressing) the vulnerability of aligned models to jailbreaking attacks.

Ryan Sheatsley, Blaine Hoak, Eric Pauley et al.

Adversarial examples, inputs designed to induce worst-case behavior in machine learning models, have been extensively studied over the past decade. Yet, our understanding of this phenomenon stems from a rather fragmented pool of knowledge; at present, there are a handful of attacks, each with disparate assumptions in threat models and incomparable definitions of optimality. In this paper, we propose a systematic approach to characterize worst-case (i.e., optimal) adversaries. We first introduce an extensible decomposition of attacks in adversarial machine learning by atomizing attack components into surfaces and travelers. With our decomposition, we enumerate over components to create 576 attacks (568 of which were previously unexplored). Next, we propose the Pareto Ensemble Attack (PEA): a theoretical attack that upper-bounds attack performance. With our new attacks, we measure performance relative to the PEA on: both robust and non-robust models, seven datasets, and three extended lp-based threat models incorporating compute costs, formalizing the Space of Adversarial Strategies. From our evaluation we find that attack performance to be highly contextual: the domain, model robustness, and threat model can have a profound influence on attack efficacy. Our investigation suggests that future studies measuring the security of machine learning should: (1) be contextualized to the domain & threat models, and (2) go beyond the handful of known attacks used today.

Yohan Beugin, Quinn Burke, Blaine Hoak et al.

Millions of consumers depend on smart camera systems to remotely monitor their homes and businesses. However, the architecture and design of popular commercial systems require users to relinquish control of their data to untrusted third parties, such as service providers (e.g., the cloud). Third parties therefore can (and in some instances have) access the video footage without the users' knowledge or consent -- violating the core tenet of user privacy. In this paper, we present CaCTUs, a privacy-preserving smart Camera system Controlled Totally by Users. CaCTUs returns control to the user; the root of trust begins with the user and is maintained through a series of cryptographic protocols, designed to support popular features, such as sharing, deleting, and viewing videos live. We show that the system can support live streaming with a latency of 2s at a frame rate of 10fps and a resolution of 480p. In so doing, we demonstrate that it is feasible to implement a performant smart-camera system that leverages the convenience of a cloud-based model while retaining the ability to control access to (private) data.

Ryan Sheatsley, Blaine Hoak, Eric Pauley et al.

Machine learning is vulnerable to adversarial examples-inputs designed to cause models to perform poorly. However, it is unclear if adversarial examples represent realistic inputs in the modeled domains. Diverse domains such as networks and phishing have domain constraints-complex relationships between features that an adversary must satisfy for an attack to be realized (in addition to any adversary-specific goals). In this paper, we explore how domain constraints limit adversarial capabilities and how adversaries can adapt their strategies to create realistic (constraint-compliant) examples. In this, we develop techniques to learn domain constraints from data, and show how the learned constraints can be integrated into the adversarial crafting process. We evaluate the efficacy of our approach in network intrusion and phishing datasets and find: (1) up to 82% of adversarial examples produced by state-of-the-art crafting algorithms violate domain constraints, (2) domain constraints are robust to adversarial examples; enforcing constraints yields an increase in model accuracy by up to 34%. We observe not only that adversaries must alter inputs to satisfy domain constraints, but that these constraints make the generation of valid adversarial examples far more challenging.

Z. Berkay Celik, Earlence Fernandes, Eric Pauley et al.

Recent advances in Internet of Things (IoT) have enabled myriad domains such as smart homes, personal monitoring devices, and enhanced manufacturing. IoT is now pervasive---new applications are being used in nearly every conceivable environment, which leads to the adoption of device-based interaction and automation. However, IoT has also raised issues about the security and privacy of these digitally augmented spaces. Program analysis is crucial in identifying those issues, yet the application and scope of program analysis in IoT remains largely unexplored by the technical community. In this paper, we study privacy and security issues in IoT that require program-analysis techniques with an emphasis on identified attacks against these systems and defenses implemented so far. Based on a study of five IoT programming platforms, we identify the key insights that result from research efforts in both the program analysis and security communities and relate the efficacy of program-analysis techniques to security and privacy issues. We conclude by studying recent IoT analysis systems and exploring their implementations. Through these explorations, we highlight key challenges and opportunities in calibrating for the environments in which IoT systems will be used.