Context is the Key: Backdoor Attacks for In-Context Learning with Vision Transformers

Due to the high cost of training, large model (LM) practitioners commonly use pretrained models downloaded from untrusted sources, which could lead to owning compromised models. In-context learning is the ability of LMs to perform multiple tasks depending on the prompt or context. This can enable new attacks, such as backdoor attacks with dynamic behavior depending on how models are prompted. In this paper, we leverage the ability of vision transformers (ViTs) to perform different tasks depending on the prompts. Then, through data poisoning, we investigate two new threats: i) task-specific backdoors where the attacker chooses a target task to attack, and only the selected task is compromised at test time under the presence of the trigger. At the same time, any other task is not affected, even if prompted with the trigger. We succeeded in attacking every tested model, achieving up to 89.90\% degradation on the target task. ii) We generalize the attack, allowing the backdoor to affect \emph{any} task, even tasks unseen during the training phase. Our attack was successful on every tested model, achieving a maximum of $13\times$ degradation. Finally, we investigate the robustness of prompts and fine-tuning as techniques for removing the backdoors from the model. We found that these methods fall short and, in the best case, reduce the degradation from 89.90\% to 73.46\%.

How to Train your Antivirus: RL-based Hardening through the Problem-Space

ML-based malware detection on dynamic analysis reports is vulnerable to both evasion and spurious correlations. In this work, we investigate a specific ML architecture employed in the pipeline of a widely-known commercial antivirus company, with the goal to harden it against adversarial malware. Adversarial training, the sole defensive technique that can confer empirical robustness, is not applicable out of the box in this domain, for the principal reason that gradient-based perturbations rarely map back to feasible problem-space programs. We introduce a novel Reinforcement Learning approach for constructing adversarial examples, a constituent part of adversarially training a model against evasion. Our approach comes with multiple advantages. It performs modifications that are feasible in the problem-space, and only those; thus it circumvents the inverse mapping problem. It also makes possible to provide theoretical guarantees on the robustness of the model against a particular set of adversarial capabilities. Our empirical exploration validates our theoretical insights, where we can consistently reach 0\% Attack Success Rate after a few adversarial retraining iterations.

Unraveling the Key of Machine Learning Solutions for Android Malware Detection

Android malware detection serves as the front line against malicious apps. With the rapid advancement of machine learning (ML), ML-based Android malware detection has attracted increasing attention due to its capability of automatically capturing malicious patterns from Android APKs. These learning-driven methods have reported promising results in detecting malware. However, the absence of an in-depth analysis of current research progress makes it difficult to gain a holistic picture of the state of the art in this area. This paper presents a comprehensive investigation to date into ML-based Android malware detection with empirical and quantitative analysis. We first survey the literature, categorizing contributions into a taxonomy based on the Android feature engineering and ML modeling pipeline. Then, we design a general-propose framework for ML-based Android malware detection, re-implement 12 representative approaches from different research communities, and evaluate them from three primary dimensions, i.e., effectiveness, robustness, and efficiency. The evaluation reveals that ML-based approaches still face open challenges and provides insightful findings like more powerful ML models are not the silver bullet for designing better malware detectors. We further summarize our findings and put forth recommendations to guide future research.

TESSERACT: Eliminating Experimental Bias in Malware Classification across Space and Time (Extended Version)

Machine learning (ML) plays a pivotal role in detecting malicious software. Despite the high F1-scores reported in numerous studies reaching upwards of 0.99, the issue is not completely solved. Malware detectors often experience performance decay due to constantly evolving operating systems and attack methods, which can render previously learned knowledge insufficient for accurate decision-making on new inputs. This paper argues that commonly reported results are inflated due to two pervasive sources of experimental bias in the detection task: spatial bias caused by data distributions that are not representative of a real-world deployment; and temporal bias caused by incorrect time splits of data, leading to unrealistic configurations. To address these biases, we introduce a set of constraints for fair experiment design, and propose a new metric, AUT, for classifier robustness in real-world settings. We additionally propose an algorithm designed to tune training data to enhance classifier performance. Finally, we present TESSERACT, an open-source framework for realistic classifier comparison. Our evaluation encompasses both traditional ML and deep learning methods, examining published works on an extensive Android dataset with 259,230 samples over a five-year span. Additionally, we conduct case studies in the Windows PE and PDF domains. Our findings identify the existence of biases in previous studies and reveal that significant performance enhancements are possible through appropriate, periodic tuning. We explore how mitigation strategies may support in achieving a more stable and better performance over time by employing multiple strategies to delay performance decay.

Adversarial Markov Games: On Adaptive Decision-Based Attacks and Defenses

Despite considerable efforts on making them robust, real-world ML-based systems remain vulnerable to decision based attacks, as definitive proofs of their operational robustness have so far proven intractable. The canonical approach in robustness evaluation calls for adaptive attacks, that is with complete knowledge of the defense and tailored to bypass it. In this study, we introduce a more expansive notion of being adaptive and show how attacks but also defenses can benefit by it and by learning from each other through interaction. We propose and evaluate a framework for adaptively optimizing black-box attacks and defenses against each other through the competitive game they form. To reliably measure robustness, it is important to evaluate against realistic and worst-case attacks. We thus augment both attacks and the evasive arsenal at their disposal through adaptive control, and observe that the same can be done for defenses, before we evaluate them first apart and then jointly under a multi-agent perspective. We demonstrate that active defenses, which control how the system responds, are a necessary complement to model hardening when facing decision-based attacks; then how these defenses can be circumvented by adaptive attacks, only to finally elicit active and adaptive defenses. We validate our observations through a wide theoretical and empirical investigation to confirm that AI-enabled adversaries pose a considerable threat to black-box ML-based systems, rekindling the proverbial arms race where defenses have to be AI-enabled too. Succinctly, we address the challenges posed by adaptive adversaries and develop adaptive defenses, thereby laying out effective strategies in ensuring the robustness of ML-based systems deployed in the real-world.

Locate and Verify: A Two-Stream Network for Improved Deepfake Detection

Deepfake has taken the world by storm, triggering a trust crisis. Current deepfake detection methods are typically inadequate in generalizability, with a tendency to overfit to image contents such as the background, which are frequently occurring but relatively unimportant in the training dataset. Furthermore, current methods heavily rely on a few dominant forgery regions and may ignore other equally important regions, leading to inadequate uncovering of forgery cues. In this paper, we strive to address these shortcomings from three aspects: (1) We propose an innovative two-stream network that effectively enlarges the potential regions from which the model extracts forgery evidence. (2) We devise three functional modules to handle the multi-stream and multi-scale features in a collaborative learning scheme. (3) Confronted with the challenge of obtaining forgery annotations, we propose a Semi-supervised Patch Similarity Learning strategy to estimate patch-level forged location annotations. Empirically, our method demonstrates significantly improved robustness and generalizability, outperforming previous methods on six benchmarks, and improving the frame-level AUC on Deepfake Detection Challenge preview dataset from 0.797 to 0.835 and video-level AUC on CelebDF$\_$v1 dataset from 0.811 to 0.847. Our implementation is available at https://github.com/sccsok/Locate-and-Verify.

DFIL: Deepfake Incremental Learning by Exploiting Domain-invariant Forgery Clues

The malicious use and widespread dissemination of deepfake pose a significant crisis of trust. Current deepfake detection models can generally recognize forgery images by training on a large dataset. However, the accuracy of detection models degrades significantly on images generated by new deepfake methods due to the difference in data distribution. To tackle this issue, we present a novel incremental learning framework that improves the generalization of deepfake detection models by continual learning from a small number of new samples. To cope with different data distributions, we propose to learn a domain-invariant representation based on supervised contrastive learning, preventing overfit to the insufficient new data. To mitigate catastrophic forgetting, we regularize our model in both feature-level and label-level based on a multi-perspective knowledge distillation approach. Finally, we propose to select both central and hard representative samples to update the replay set, which is beneficial for both domain-invariant representation learning and rehearsal-based knowledge preserving. We conduct extensive experiments on four benchmark datasets, obtaining the new state-of-the-art average forgetting rate of 7.01 and average accuracy of 85.49 on FF++, DFDC-P, DFD, and CDF2. Our code is released at https://github.com/DeepFakeIL/DFIL.

Symmetry-Preserving Program Representations for Learning Code Semantics

Large Language Models (LLMs) have shown promise in automated program reasoning, a crucial aspect of many security tasks. However, existing LLM architectures for code are often borrowed from other domains like natural language processing, raising concerns about their generalization and robustness to unseen code. A key generalization challenge is to incorporate the knowledge of code semantics, including control and data flow, into the LLM architectures. Drawing inspiration from examples of convolution layers exploiting translation symmetry, we explore how code symmetries can enhance LLM architectures for program analysis and modeling. We present a rigorous group-theoretic framework that formally defines code symmetries as semantics-preserving transformations and provides techniques for precisely reasoning about symmetry preservation within LLM architectures. Using this framework, we introduce a novel variant of self-attention that preserves program symmetries, demonstrating its effectiveness in generalization and robustness through detailed experimental evaluations across different binary and source code analysis tasks. Overall, our code symmetry framework offers rigorous and powerful reasoning techniques that can guide the future development of specialized LLMs for code and advance LLM-guided program reasoning tasks.

Jigsaw Puzzle: Selective Backdoor Attack to Subvert Malware Classifiers

Malware classifiers are subject to training-time exploitation due to the need to regularly retrain using samples collected from the wild. Recent work has demonstrated the feasibility of backdoor attacks against malware classifiers, and yet the stealthiness of such attacks is not well understood. In this paper, we investigate this phenomenon under the clean-label setting (i.e., attackers do not have complete control over the training or labeling process). Empirically, we show that existing backdoor attacks in malware classifiers are still detectable by recent defenses such as MNTD. To improve stealthiness, we propose a new attack, Jigsaw Puzzle (JP), based on the key observation that malware authors have little to no incentive to protect any other authors' malware but their own. As such, Jigsaw Puzzle learns a trigger to complement the latent patterns of the malware author's samples, and activates the backdoor only when the trigger and the latent pattern are pieced together in a sample. We further focus on realizable triggers in the problem space (e.g., software code) using bytecode gadgets broadly harvested from benign software. Our evaluation confirms that Jigsaw Puzzle is effective as a backdoor, remains stealthy against state-of-the-art defenses, and is a threat in realistic settings that depart from reasoning about feature-space only attacks. We conclude by exploring promising approaches to improve backdoor defenses.

Universal Adversarial Perturbations for Malware

Machine learning classification models are vulnerable to adversarial examples -- effective input-specific perturbations that can manipulate the model's output. Universal Adversarial Perturbations (UAPs), which identify noisy patterns that generalize across the input space, allow the attacker to greatly scale up the generation of these adversarial examples. Although UAPs have been explored in application domains beyond computer vision, little is known about their properties and implications in the specific context of realizable attacks, such as malware, where attackers must reason about satisfying challenging problem-space constraints. In this paper, we explore the challenges and strengths of UAPs in the context of malware classification. We generate sequences of problem-space transformations that induce UAPs in the corresponding feature-space embedding and evaluate their effectiveness across threat models that consider a varying degree of realistic attacker knowledge. Additionally, we propose adversarial training-based mitigations using knowledge derived from the problem-space transformations, and compare against alternative feature-space defenses. Our experiments limit the effectiveness of a white box Android evasion attack to ~20 % at the cost of 3 % TPR at 1 % FPR. We additionally show how our method can be adapted to more restrictive application domains such as Windows malware. We observe that while adversarial training in the feature space must deal with large and often unconstrained regions, UAPs in the problem space identify specific vulnerabilities that allow us to harden a classifier more effectively, shifting the challenges and associated cost of identifying new universal adversarial transformations back to the attacker.