Fnu Suya

Kunlin Cai, Rui Song, Jinghuai Zhang et al.

Video generation models are increasingly used as world simulators for tasks like driving and robotic manipulation. What matters in these settings is not whether a single video looks right, but whether the model's output changes when its input changes. We test this by giving a model two prompts describing the same scene with one physical detail varied, and checking whether the two videos diverge the way physics predicts. The wording difference between the prompts is small by design, since only one variable is changed, but the correct physical difference is not. A model that misses this can still produce two videos that each look plausible individually, and existing benchmarks score videos one at a time and cannot detect this failure. We introduce What-If World, 319 such prompt pairs built on real frames from nuScenes and DROID, organized by a taxonomy of six physical variables shared across driving and manipulation. Each pair is scored with APEO, a four-part rubric checking whether each video follows its prompt (Adherence), is physically consistent (Physics), preserves the shared scene (Environment), and ends in the correct difference (Outcome). Across nine state-of-the-art models, no system exceeds 52% on the paired score, and open-source models cluster near 28%. Every model tested fails on a large fraction of causal interventions, indicating substantial room before these models can reliably support action-conditioned simulation or model-based planning. Where models do score well, performance appears to track the visual prominence of the intervention rather than the tractability of its underlying physics. Some visually subtle interventions score as low as 14.2%, while visually pronounced ones reach 40.4%.

Jinghuai Zhang, Pengyue Yu, Zhexiao Lin et al.

Image-based Retrieval-Augmented Generation (IRAG) conditions a frozen generator on reference images retrieved from an external database, supporting both text-to-image (T2I) and question answering (Q&A) tasks. Because these databases are opaque and web-scraped, copyright holders need ways to audit whether specific images appear in them. While prior work employs membership inference attacks (MIAs) to audit uni-modal, text-based RAG, they fail to transfer to IRAG due to two key challenges. First, cross-modal retrieval: text-RAG MIAs force retrieval of the target passage by injecting its content into the query, which is unavailable in IRAG since images cannot be embedded into text queries; even accurate image captions fail to bridge the modality gap. Second, discriminative signal extraction: text-RAG MIAs extract membership signals by prompting the generator to answer multiple questions over the target passage, whereas T2I generators in IRAG produce images rather than follow Q&A commands. To fill this gap, we introduce the first MIA tailored to IRAG, ImageAuditor, which decomposes each attack query into a retrieval segment and an extraction segment, enabling dedicated optimization for each challenge. For retrieval, we propose Reward-Guided Policy Optimization (RGPO), which updates a stochastic policy from reward-ranked candidates to navigate the cross-modal embedding landscape and admits finite-sample optimality guarantees to balance exploration and exploitation. For extraction, we analyze the distribution of the MIA score to guide the co-design of the prompting strategy and scoring rule, and derive task-specific instantiations for T2I and Q&A tasks. We aggregate signals across queries via K-means clustering for reliable membership decisions. Across various IRAG systems, ImageAuditor exceeds 80% AUROC with only four queries per audited image and remains robust across diverse settings.

Jinghuai Zhang, Yetian He, Kunlin Cai et al.

Model merging composes specialized capabilities into a single LLM by aggregating task vectors sourced from unverified public platforms, exposing a critical supply-chain attack surface: Because any malicious behavior can be encoded into a task vector, and merging grants third-party vectors direct write access to model weights, an attacker-provided task vector can enable or amplify diverse downstream threats. Prior work studies only backdoor attacks against model merging for classifiers using static arithmetic heuristics, which fail to effectively handle diverse attacks on generative LLMs for three reasons. (i) LLMs rely on autoregressive decoding, where the minor parameter drift introduced by merging compounds across tokens and rapidly degrades the attack. (ii) Attackers have no knowledge of the victim's merging configurations, causing a static attack vector optimized in isolation to be easily diluted or destroyed. (iii) Practical threat induction must generalize to attack prompts unseen during optimization, which static vectors cannot adequately encode. We present RogueMerge, the first principled, unified framework that addresses all three challenges. To handle autoregressive generation, we replace static arithmetic with a joint optimization that explicitly enforces attack success after merging. To handle unknown merging settings, we formulate attack injection as a stochastic min-max problem and solve it via meta-learning-style simulation. To generalize across heterogeneous attack prompts, we employ distributionally robust optimization and derive a tractable first-order Taylor approximation at LLM scale, with a provable error bound. Across four threats, six merging algorithms, and over 170 merged LLMs, RogueMerge consistently outperforms existing attacks. It also remains stable across diverse merging settings and resists standard defenses.

Yulong Tian, Fnu Suya, Anshuman Suri et al.

Transfer learning is a popular method for tuning pretrained (upstream) models for different downstream tasks using limited data and computational resources. We study how an adversary with control over an upstream model used in transfer learning can conduct property inference attacks on a victim's tuned downstream model. For example, to infer the presence of images of a specific individual in the downstream training set. We demonstrate attacks in which an adversary can manipulate the upstream model to conduct highly effective and specific property inference attacks (AUC score $> 0.9$), without incurring significant performance loss on the main task. The main idea of the manipulation is to make the upstream model generate activations (intermediate features) with different distributions for samples with and without a target property, thus enabling the adversary to distinguish easily between downstream models trained with and without training examples that have the target property. Our code is available at https://github.com/yulongt23/Transfer-Inference.

Tingwei Zhang, Fnu Suya, Rishi Jha et al.

Hubness is a phenomenon in high-dimensional vector spaces where a point from the natural distribution is unusually close to many other points. This is a well-known problem in information retrieval that causes some items to accidentally (and incorrectly) appear relevant to many queries. In this paper, we investigate how attackers can exploit hubness to turn any image or audio input in a multi-modal retrieval system into an adversarial hub. Adversarial hubs can be used to inject universal adversarial content (e.g., spam) that will be retrieved in response to thousands of different queries, and also for targeted attacks on queries related to specific, attacker-chosen concepts. We present a method for creating adversarial hubs and evaluate the resulting hubs on benchmark multi-modal retrieval datasets and an image-to-image retrieval system implemented by Pinecone, a popular vector database. For example, in text-caption-to-image retrieval, a single adversarial hub, generated using 100 random queries, is retrieved as the top-1 most relevant image for more than 21,000 out of 25,000 test queries (by contrast, the most common natural hub is the top-1 response to only 102 queries), demonstrating the strong generalization capabilities of adversarial hubs. We also investigate whether techniques for mitigating natural hubness can also mitigate adversarial hubs, and show that they are not effective against hubs that target queries related to specific concepts.

Shahinul Hoque, Jinghuai Zhang, Jinyuan Sun et al.

Per-token billing is now the standard pricing model for commercial large language models (LLMs), so the honesty of reported token counts directly affects what users pay. We show that this kind of billing is hard to audit by design: providers hide the model, the tokenizer, and the execution to protect their IP, mitigate jailbreaks, and preserve user privacy, which means an auditor can only inspect proofs the provider supplies. The audit therefore reduces to a consistency check on the provider's own reports. We call this a trust paradox: every audit must trust some artifact, but current frameworks trust exactly the ones a provider has the strongest reason to manipulate. We study three recent token auditing frameworks and show that a provider with ordinary commercial capabilities can systematically inflate billed token counts. In the most permissive setting, hidden reasoning usage can be inflated by 1,469% on average without detection. At current frontier reasoning prices, that turns a \$100 honest bill into roughly a \$1,569 bill on the same query. Even when the user can see the full reasoning string, tokenization ambiguity alone still allows 50.85% over-reporting below the detection threshold. These results suggest the problem is not in any specific auditor but in any audit whose evidence comes from the audited party. Restoring honest billing will require verification that ties reported token counts to evidence the provider does not control, such as trusted execution attestation, cryptographic proofs of inference, or third-party re-execution.

Fnu Suya, Anshuman Suri, Tingwei Zhang et al.

Numerous works study black-box attacks on image classifiers. However, these works make different assumptions on the adversary's knowledge and current literature lacks a cohesive organization centered around the threat model. To systematize knowledge in this area, we propose a taxonomy over the threat space spanning the axes of feedback granularity, the access of interactive queries, and the quality and quantity of the auxiliary data available to the attacker. Our new taxonomy provides three key insights. 1) Despite extensive literature, numerous under-explored threat spaces exist, which cannot be trivially solved by adapting techniques from well-explored settings. We demonstrate this by establishing a new state-of-the-art in the less-studied setting of access to top-k confidence scores by adapting techniques from well-explored settings of accessing the complete confidence vector, but show how it still falls short of the more restrictive setting that only obtains the prediction label, highlighting the need for more research. 2) Identification the threat model of different attacks uncovers stronger baselines that challenge prior state-of-the-art claims. We demonstrate this by enhancing an initially weaker baseline (under interactive query access) via surrogate models, effectively overturning claims in the respective paper. 3) Our taxonomy reveals interactions between attacker knowledge that connect well to related areas, such as model inversion and extraction attacks. We discuss how advances in other areas can enable potentially stronger black-box attacks. Finally, we emphasize the need for a more realistic assessment of attack success by factoring in local attack runtime. This approach reveals the potential for certain attacks to achieve notably higher success rates and the need to evaluate attacks in diverse and harder settings, highlighting the need for better selection criteria.

Abdullah Al Nomaan Nafi, Fnu Suya, Swarup Bhunia et al.

Jailbreak attacks expose a persistent gap between the intended safety behavior of aligned large language models and their behavior under adversarial prompting. Existing automated methods are increasingly effective but each commits to a single attack family (e.g., one refinement loop, one tree search, one mutation space, or one strategy library) and no single family dominates: the best-performing method shifts across target models and harm categories, suggesting complementary strengths that per-prompt composition could exploit. We introduce LASH (LLM Adaptive Semantic Hybridization), a black-box framework that treats outputs from multiple base attacks as reusable seed prompts and adaptively composes them for each target request. Given a seed pool, LASH searches over seed subsets and softmax-normalized mixture weights; a composition module synthesizes a single candidate prompt, and a derivative-free genetic optimizer updates the weights using black-box target feedback and a two-stage fitness function combining keyword-based refusal detection with LLM-judge scoring. On JailbreakBench, which contains 100 harmful prompts across 10 categories, we evaluate LASH on six common target models. LASH achieves an average attack success rate of 84.5% under keyword-based evaluation and 74.5% under two-stage evaluation, where responses are first filtered for refusals and then scored by an LLM judge for whether they substantively fulfill the original harmful request. LASH outperforms five state-of-the-art baselines on both metrics with only 30 mean target queries. LASH also remains competitive under three defense mechanisms and induces more success-like internal representations. These results suggest that adaptive composition across heterogeneous jailbreak strategies is a promising direction for black-box red-teaming.

Fnu Suya, Xiao Zhang, Yuan Tian et al.

We study indiscriminate poisoning for linear learners where an adversary injects a few crafted examples into the training data with the goal of forcing the induced model to incur higher test error. Inspired by the observation that linear learners on some datasets are able to resist the best known attacks even without any defenses, we further investigate whether datasets can be inherently robust to indiscriminate poisoning attacks for linear learners. For theoretical Gaussian distributions, we rigorously characterize the behavior of an optimal poisoning attack, defined as the poisoning strategy that attains the maximum risk of the induced model at a given poisoning budget. Our results prove that linear learners can indeed be robust to indiscriminate poisoning if the class-wise data distributions are well-separated with low variance and the size of the constraint set containing all permissible poisoning points is also small. These findings largely explain the drastic variation in empirical attack performance of the state-of-the-art poisoning attacks on linear learners across benchmark datasets, making an important initial step towards understanding the underlying reasons some learning tasks are vulnerable to data poisoning attacks.

Evan Rose, Fnu Suya, David Evans

Machine learning is susceptible to poisoning attacks, in which an attacker controls a small fraction of the training data and chooses that data with the goal of inducing some behavior unintended by the model developer in the trained model. We consider a realistic setting in which the adversary with the ability to insert a limited number of data points attempts to control the model's behavior on a specific subpopulation. Inspired by previous observations on disparate effectiveness of random label-flipping attacks on different subpopulations, we investigate the properties that can impact the effectiveness of state-of-the-art poisoning attacks against different subpopulations. For a family of 2-dimensional synthetic datasets, we empirically find that dataset separability plays a dominant role in subpopulation vulnerability for less separable datasets. However, well-separated datasets exhibit more dependence on individual subpopulation properties. We further discover that a crucial subpopulation property is captured by the difference in loss on the clean dataset between the clean model and a target model that misclassifies the subpopulation, and a subpopulation is much easier to attack if the loss difference is small. This property also generalizes to high-dimensional benchmark datasets. For the Adult benchmark dataset, we show that we can find semantically-meaningful subpopulation properties that are related to the susceptibilities of a selected group of subpopulations. The results in this paper are accompanied by a fully interactive web-based visualization of subpopulation poisoning attacks found at https://uvasrg.github.io/visualizing-poisoning

Sanskar Amgain, Daniel Lobo, Atri Chatterjee et al.

The growing use of third-party hardware accelerators (e.g., FPGAs, ASICs) for deep neural networks (DNNs) introduces new security vulnerabilities. Conventional model-level backdoor attacks, which only poison a model's weights to misclassify inputs with a specific trigger, are often detectable because the entire attack logic is embedded within the model (i.e., software), creating a traceable layer-by-layer activation path. This paper introduces the HArdware-Model Logically Combined Attack (HAMLOCK), a far stealthier threat that distributes the attack logic across the hardware-software boundary. The software (model) is now only minimally altered by tuning the activations of few neurons to produce uniquely high activation values when a trigger is present. A malicious hardware Trojan detects those unique activations by monitoring the corresponding neurons' most significant bit or the 8-bit exponents and triggers another hardware Trojan to directly manipulate the final output logits for misclassification. This decoupled design is highly stealthy, as the model itself contains no complete backdoor activation path as in conventional attacks and hence, appears fully benign. Empirically, across benchmarks like MNIST, CIFAR10, GTSRB, and ImageNet, HAMLOCK achieves a near-perfect attack success rate with a negligible clean accuracy drop. More importantly, HAMLOCK circumvents the state-of-the-art model-level defenses without any adaptive optimization. The hardware Trojan is also undetectable, incurring area and power overheads as low as 0.01%, which is easily masked by process and environmental noise. Our findings expose a critical vulnerability at the hardware-software interface, demanding new cross-layer defenses against this emerging threat.

Abdullah Al Nomaan Nafi, Habibur Rahaman, Zafaryab Haider et al.

Numerous techniques have been proposed for generating adversarial examples in white-box settings under strict Lp-norm constraints. However, such norm-bounded examples often fail to align well with human perception, and only recently have a few methods begun specifically exploring perceptually aligned adversarial examples. Moreover, it remains unclear whether insights from Lp-constrained attacks can be effectively leveraged to improve perceptual efficacy. In this paper, we introduce DAASH, a fully differentiable meta-attack framework that generates effective and perceptually aligned adversarial examples by strategically composing existing Lp-based attack methods. DAASH operates in a multi-stage fashion: at each stage, it aggregates candidate adversarial examples from multiple base attacks using learned, adaptive weights and propagates the result to the next stage. A novel meta-loss function guides this process by jointly minimizing misclassification loss and perceptual distortion, enabling the framework to dynamically modulate the contribution of each base attack throughout the stages. We evaluate DAASH on adversarially trained models across CIFAR-10, CIFAR-100, and ImageNet. Despite relying solely on Lp-constrained based methods, DAASH significantly outperforms state-of-the-art perceptual attacks such as AdvAD -- achieving higher attack success rates (e.g., 20.63\% improvement) and superior visual quality, as measured by SSIM, LPIPS, and FID (improvements $\approx$ of 11, 0.015, and 5.7, respectively). Furthermore, DAASH generalizes well to unseen defenses, making it a practical and strong baseline for evaluating robustness without requiring handcrafted adaptive attacks for each new defense.

Yulong Tian, Fnu Suya, Fengyuan Xu et al.

In a backdoor attack on a machine learning model, an adversary produces a model that performs well on normal inputs but outputs targeted misclassifications on inputs containing a small trigger pattern. Model compression is a widely-used approach for reducing the size of deep learning models without much accuracy loss, enabling resource-hungry models to be compressed for use on resource-constrained devices. In this paper, we study the risk that model compression could provide an opportunity for adversaries to inject stealthy backdoors. We design stealthy backdoor attacks such that the full-sized model released by adversaries appears to be free from backdoors (even when tested using state-of-the-art techniques), but when the model is compressed it exhibits highly effective backdoors. We show this can be done for two common model compression techniques -- model pruning and model quantization. Our findings demonstrate how an adversary may be able to hide a backdoor as a compression artifact, and show the importance of performing security tests on the models that will actually be deployed not their precompressed version.

Fnu Suya, Saeed Mahloujifar, Anshuman Suri et al.

In a poisoning attack, an adversary with control over a small fraction of the training data attempts to select that data in a way that induces a corrupted model that misbehaves in favor of the adversary. We consider poisoning attacks against convex machine learning models and propose an efficient poisoning attack designed to induce a specified model. Unlike previous model-targeted poisoning attacks, our attack comes with provable convergence to {\it any} attainable target classifier. The distance from the induced classifier to the target classifier is inversely proportional to the square root of the number of poisoning points. We also provide a lower bound on the minimum number of poisoning points needed to achieve a given target classifier. Our method uses online convex optimization, so finds poisoning points incrementally. This provides more flexibility than previous attacks which require a priori assumption about the number of poisoning points. Our attack is the first model-targeted poisoning attack that provides provable convergence for convex models, and in our experiments, it either exceeds or matches state-of-the-art attacks in terms of attack success rate and distance to the target model.

Jihong Wang, Minnan Luo, Fnu Suya et al.

Recent studies have shown that graph convolution networks (GCNs) are vulnerable to carefully designed attacks, which aim to cause misclassification of a specific node on the graph with unnoticeable perturbations. However, a vast majority of existing works cannot handle large-scale graphs because of their high time complexity. Additionally, existing works mainly focus on manipulating existing nodes on the graph, while in practice, attackers usually do not have the privilege to modify information of existing nodes. In this paper, we develop a more scalable framework named Approximate Fast Gradient Sign Method (AFGSM) which considers a more practical attack scenario where adversaries can only inject new vicious nodes to the graph while having no control over the original graph. Methodologically, we provide an approximation strategy to linearize the model we attack and then derive an approximate closed-from solution with a lower time cost. To have a fair comparison with existing attack methods that manipulate the original graph, we adapt them to the new attack scenario by injecting vicious nodes. Empirical experimental results show that our proposed attack method can significantly reduce the classification accuracy of GCNs and is much faster than existing methods without jeopardizing the attack performance.

Fnu Suya, Jianfeng Chi, David Evans et al.

We study adversarial examples in a black-box setting where the adversary only has API access to the target model and each query is expensive. Prior work on black-box adversarial examples follows one of two main strategies: (1) transfer attacks use white-box attacks on local models to find candidate adversarial examples that transfer to the target model, and (2) optimization-based attacks use queries to the target model and apply optimization techniques to search for adversarial examples. We propose hybrid attacks that combine both strategies, using candidate adversarial examples from local models as starting points for optimization-based attacks and using labels learned in optimization-based attacks to tune local models for finding transfer candidates. We empirically demonstrate on the MNIST, CIFAR10, and ImageNet datasets that our hybrid attack strategy reduces cost and improves success rates. We also introduce a seed prioritization strategy which enables attackers to focus their resources on the most promising seeds. Combining hybrid attacks with our seed prioritization strategy enables batch attacks that can reliably find adversarial examples with only a handful of queries.

Fnu Suya, Yuan Tian, David Evans et al.

We study black-box attacks on machine learning classifiers where each query to the model incurs some cost or risk of detection to the adversary. We focus explicitly on minimizing the number of queries as a major objective. Specifically, we consider the problem of attacking machine learning classifiers subject to a budget of feature modification cost while minimizing the number of queries, where each query returns only a class and confidence score. We describe an approach that uses Bayesian optimization to minimize the number of queries, and find that the number of queries can be reduced to approximately one tenth of the number needed through a random strategy for scenarios where the feature modification cost budget is low.