Automation platforms such as GitHub Actions and n8n are increasingly adopting so-called agentic workflows, which integrate Large Language Model (LLM) agents for tasks such as code review and data synchronization. While bringing convenience for developers, this integration exposes a new risk: An adversary may control and craft certain inputs, such as GitHub issue comments, to manipulate the LLM agent for unwanted actions, such as credential exfiltration and arbitrary command execution. To our knowledge, no prior academic work has studied such a risk in agentic workflows. In this paper, we design the first detection and exploitation framework, called JAW, to hijack agentic workflows hosted on automation platforms via a novel approach called Context-Grounded Evolution. Our key idea is to evolve agentic workflow inputs under the contexts derived from hybrid program analysis for hijacking purposes. Specifically, JAW generates agentic workflow contexts through three analyses: (i) static path-feasibility analysis to identify feasible agent-invocation paths and the input constraints required to trigger them, (ii) dynamic prompt-provenance analysis to determine how that input is transformed and embedded into the LLM context, and (iii) capability analysis to identify the actions and restrictions available to the agent at runtime. Our evaluation of JAW on GitHub workflows and n8n templates showed that 4714 GitHub workflows and eight n8n templates can be successfully hijacked, for example, to leak user credentials. Our findings span 15 widely-used GitHub Actions, including official GitHub Actions for Claude Code, Gemini CLI, Qwen CLI, and Cursor CLI, and two official n8n nodes. We responsibly disclosed all findings to the affected vendors and received many acknowledgements, fixes, and bug bounties, notably from GitHub, Google, and Anthropic.

Safe agents can fail together. Multi-agent LLM systems (MAS) move information, state, decisions, and authority across principal boundaries, creating failures that local checks may miss. Without an execution-level view, a multi-agent setting can easily be mistaken for evidence of a genuinely multi-agent security effect. We thus systematize MAS security through an execution-centered analysis of 197 works, covering six interaction interfaces, four adversary positions, seven system-level risks, and eight recurring attack paths. We introduce an A-I-R framework that organizes attacks by adversary position, interaction interface, and resulting system-level risk, unifying otherwise fragmented attack mechanisms across MAS. We organize defenses through a five-part contract covering path target, observation, intervention, trust boundary, and recovery, and identify path closure and recovery as key challenges. We audit 44 evaluation and benchmark works and identify open challenges in isolating interaction effects, designing comparable and diagnostic metrics, supporting reuse across MAS designs, and evaluating open-system operation. Together, these findings motivate an interaction-aware view of MAS security: trace attacks end to end, test whether defenses close those paths, and evaluate system-level effects with appropriate counterfactuals.

With the widespread availability of pretrained Large Language Models (LLMs) and their training datasets, concerns about the security risks associated with their usage has increased significantly. One of these security risks is the threat of LLM poisoning attacks where an attacker modifies some part of the LLM training process to cause the LLM to behave in a malicious way. As an emerging area of research, the current frameworks and terminology for LLM poisoning attacks are derived from earlier classification poisoning literature and are not fully equipped for generative LLM settings. We conduct a systematic review of published LLM poisoning attacks to clarify the security implications and address inconsistencies in terminology across the literature. We propose a comprehensive poisoning threat model applicable to categorize a wide range of LLM poisoning attacks. The poisoning threat model includes four poisoning attack specifications that define the logistics and manipulation strategies of an attack as well as six poisoning metrics used to measure key characteristics of an attack. Under our proposed framework, we organize our discussion of published LLM poisoning literature along four critical dimensions of LLM poisoning attacks: concept poisons, stealthy poisons, persistent poisons, and poisons for unique tasks, to better understand the current landscape of security risks.

Machine unlearning methods take a model trained on a dataset and a forget set, then attempt to produce a model as if it had only been trained on the examples not in the forget set. We empirically show that an adversary is able to distinguish between a mirror model (a control model produced by retraining without the data to forget) and a model produced by an unlearning method across representative unlearning methods from the literature. We build distinguishing algorithms based on evaluation scores in the literature (i.e. membership inference scores) and Kullback-Leibler divergence. We propose a strong formal definition for machine unlearning called computational unlearning. Computational unlearning is defined as the inability for an adversary to distinguish between a mirror model and a model produced by an unlearning method. If the adversary cannot guess better than random (except with negligible probability), then we say that an unlearning method achieves computational unlearning. Our computational unlearning definition provides theoretical structure to prove unlearning feasibility results. For example, our computational unlearning definition immediately implies that there are no deterministic computational unlearning methods for entropic learning algorithms. We also explore the relationship between differential privacy (DP)-based unlearning methods and computational unlearning, showing that DP-based approaches can satisfy computational unlearning at the cost of an extreme utility collapse. These results demonstrate that current methodology in the literature fundamentally falls short of achieving computational unlearning. We conclude by identifying several open questions for future work.

Many online communications systems use perceptual hash matching systems to detect illicit files in user content. These systems employ specialized perceptual hash functions such as Microsoft's PhotoDNA or Facebook's PDQ to produce a compact digest of an image file that can be approximately compared to a database of known illicit-content digests. Recently, several proposals have suggested that hash-based matching systems be incorporated into client-side and end-to-end encrypted (E2EE) systems: in these designs, files that register as illicit content will be reported to the provider, while the remaining content will be sent confidentially. By using perceptual hashing to determine confidentiality guarantees, this new setting significantly changes the function of existing perceptual hashing - thus motivating the need to evaluate these functions from an adversarial perspective, using their perceptual capabilities against them. For example, an attacker may attempt to trigger a match on innocuous, but politically-charged, content in an attempt to stifle speech. In this work we develop threat models for perceptual hashing algorithms in an adversarial setting, and present attacks against the two most widely deployed algorithms: PhotoDNA and PDQ. Our results show that it is possible to efficiently generate targeted second-preimage attacks in which an attacker creates a variant of some source image that matches some target digest. As a complement to this main result, we also further investigate the production of images that facilitate detection avoidance attacks, continuing a recent investigation of Jain et al. Our work shows that existing perceptual hash functions are likely insufficiently robust to survive attacks on this new setting.

Measuring and attributing greenhouse gas (GHG) emissions remains a challenging problem as the world strives toward meeting emissions reductions targets. As a significant portion of total global emissions, the road transportation sector represents an enormous challenge for estimating and tracking emissions at a global scale. To meet this challenge, we have developed a hybrid approach for estimating road transportation emissions that combines the strengths of machine learning and satellite imagery with localized emissions factors data to create an accurate, globally scalable, and easily configurable GHG monitoring framework.

Membership inference (MI) attacks are more diverse in a Federated Learning (FL) setting, because an adversary may be either an FL client, a server, or an external attacker. Existing defenses against MI attacks rely on perturbations to either the model's output predictions or the training process. However, output perturbations are ineffective in an FL setting, because a malicious server can access the model without output perturbation while training perturbations struggle to achieve a good utility. This paper proposes a novel defense, called CIP, to fortify FL against MI attacks via a client-level input perturbation during training and inference procedures. The key insight is to shift each client's local data distribution via a personalized perturbation to get a shifted model. CIP achieves a good balance between privacy and utility. Our evaluation shows that CIP causes accuracy to drop at most 0.7% while reducing attacks to random guessing.

Whilst adversarial attack detection has received considerable attention, it remains a fundamentally challenging problem from two perspectives. First, while threat models can be well-defined, attacker strategies may still vary widely within those constraints. Therefore, detection should be considered as an open-set problem, standing in contrast to most current detection approaches. These methods take a closed-set view and train binary detectors, thus biasing detection toward attacks seen during detector training. Second, limited information is available at test time and typically confounded by nuisance factors including the label and underlying content of the image. We address these challenges via a novel strategy based on random sub-space analysis. We present a technique that utilizes properties of random projections to characterize the behavior of clean and adversarial examples across a diverse set of subspaces. The self-consistency (or inconsistency) of model activations is leveraged to discern clean from adversarial examples. Performance evaluations demonstrate that our technique (AUC ∈ [0.92, 0.98]) outperforms competing detection strategies (AUC ∈ [0.30, 0.79]), while remaining truly agnostic to the attack strategy (for both targeted/untargeted attacks). It also requires significantly less calibration data (composed only of clean examples) than competing approaches to achieve this performance.

Training reinforcement learning agents that continually learn across multiple environments is a challenging problem. This is made more difficult by a lack of reproducible experiments and standard metrics for comparing different continual learning approaches. To address this, we present TELLA, a tool for the Test and Evaluation of Lifelong Learning Agents. TELLA provides specified, reproducible curricula to lifelong learning agents while logging detailed data for evaluation and standardized analysis. Researchers can define and share their own curricula over various learning environments or run against a curriculum created under the DARPA Lifelong Learning Machines (L2M) Program.

Despite groundbreaking progress in reinforcement learning for robotics, gameplay, and other complex domains, major challenges remain in applying reinforcement learning to the evolving, open-world problems often found in critical application spaces. Reinforcement learning solutions tend to generalize poorly when exposed to new tasks outside of the data distribution they are trained on, prompting an interest in continual learning algorithms. In tandem with research on continual learning algorithms, there is a need for challenge environments, carefully designed experiments, and metrics to assess research progress. We address the latter need by introducing a framework for continual reinforcement-learning development and assessment using Lifelong Learning Explorer (L2Explorer), a new, Unity-based, first-person 3D exploration environment that can be continuously reconfigured to generate a range of tasks and task variants structured into complex and evolving evaluation curricula. In contrast to procedurally generated worlds with randomized components, we have developed a systematic approach to defining curricula in response to controlled changes with accompanying metrics to assess transfer, performance recovery, and data efficiency. Taken together, the L2Explorer environment and evaluation approach provides a framework for developing future evaluation methodologies in open-world settings and rigorously evaluating approaches to lifelong learning.

Searching for small objects in large images is a task that is both challenging for current deep learning systems and important in numerous real-world applications, such as remote sensing and medical imaging. Thorough scanning of very large images is computationally expensive, particularly at resolutions sufficient to capture small objects. The smaller an object of interest, the more likely it is to be obscured by clutter or otherwise deemed insignificant. We examine these issues in the context of two complementary problems: closed-set object detection and open-set target search. First, we present a method for predicting pixel-level objectness from a low resolution gist image, which we then use to select regions for performing object detection locally at high resolution. This approach has the benefit of not being fixed to a predetermined grid, thereby requiring fewer costly high-resolution glimpses than existing methods. Second, we propose a novel strategy for open-set visual search that seeks to find all instances of a target class which may be previously unseen and is defined by a single image. We interpret both detection problems through a probabilistic, Bayesian lens, whereby the objectness maps produced by our method serve as priors in a maximum-a-posteriori approach to the detection step. We evaluate the end-to-end performance of both the combination of our patch selection strategy with this target search approach and the combination of our patch selection strategy with standard object detection methods. Both elements of our approach are seen to significantly outperform baseline strategies.

We focus on the development of effective adversarial patch attacks and – for the first time – jointly address the antagonistic objectives of attack success and obtrusiveness via the design of novel semi-transparent patches. This work is motivated by our pursuit of a systematic performance analysis of patch attack robustness with regard to geometric transformations. Specifically, we first elucidate a) key factors underpinning patch attack success and b) the impact of distributional shift between training and testing/deployment when cast under the Expectation over Transformation (EoT) formalism. By focusing our analysis on three principal classes of transformations (rotation, scale, and location), our findings provide quantifiable insights into the design of effective patch attacks and demonstrate that scale, among all factors, significantly impacts patch attack success. Working from these findings, we then focus on addressing how to overcome the principal limitations of scale for the deployment of attacks in real physical settings: namely the obtrusiveness of large patches. Our strategy is to turn to the novel design of irregularly-shaped, semi-transparent partial patches which we construct via a new optimization process that jointly addresses the antagonistic goals of mitigating obtrusiveness and maximizing effectiveness. Our study – we hope – will help encourage more focus in the community on the issues of obtrusiveness, scale, and success in patch attacks.

With deep neural networks (DNNs) being used increasingly in many applications, it is critical to improve our understanding of their failure modes and potential mitigations. A Johns Hopkins University Applied Physics Laboratory (APL) team successfully inserted a backdoor (train-time attack) into a common object detection model. In conjunction with this research, they developed a principled methodology to evaluate patch attacks (test-time attacks) and the factors impacting their success. Their approach enabled the creation of a novel optimization framework for the first-ever design of semitransparent patches that can overcome scale limitations while retaining desirable factors with regard to deployment and detectability.

The Active Sensing Testbed (AST) is a novel framework for research in machine perception and world-view reasoning. The AST supports exploratory development of perception systems that can build internal models of the world by combining multi-view and multi-modal analytics, utilize these models to form hypotheses about a scene, and potentially take action to fill in gaps in knowledge or make predictions about future world states. As a modular software framework, the AST is intended to lower the barrier to entry for researchers and developers in applying state-of-the-art computer vision techniques to real-world problems.

In this paper, we introduce the TrojAI software framework, an open source set of Python tools capable of generating triggered (poisoned) datasets and associated deep learning (DL) models with trojans at scale. We utilize the developed framework to generate a large set of trojaned MNIST classifiers, as well as demonstrate the capability to produce a trojaned reinforcement-learning model using vector observations. Results on MNIST show that the nature of the trigger, training batch size, and dataset poisoning percentage all affect successful embedding of trojans. We test Neural Cleanse against the trojaned MNIST models and successfully detect anomalies in the trained models approximately $18\%$ of the time. Our experiments and workflow indicate that the TrojAI software framework will enable researchers to easily understand the effects of various configurations of the dataset and training hyperparameters on the generated trojaned deep learning model, and can be used to rapidly and comprehensively test new trojan detection methods.

Searching for small objects in large images is currently challenging for deep learning systems, but is a task with numerous applications including remote sensing and medical imaging. Thorough scanning of very large images is computationally expensive, particularly at resolutions sufficient to capture small objects. The smaller an object of interest, the more likely it is to be obscured by clutter or otherwise deemed insignificant. We examine these issues in the context of two complementary problems: closed-set object detection and open-set target search. First, we present a method for predicting pixel-level objectness from a low resolution gist image, which we then use to select regions for subsequent evaluation at high resolution. This approach has the benefit of not being fixed to a predetermined grid, allowing fewer costly high-resolution glimpses than existing methods. Second, we propose a novel strategy for open-set visual search that seeks to find all objects in an image of the same class as a given target reference image. We interpret both detection problems through a probabilistic, Bayesian lens, whereby the objectness maps produced by our method serve as priors in a maximum-a-posteriori approach to the detection step. We evaluate the end-to-end performance of both the combination of our patch selection strategy with this target search approach and the combination of our patch selection strategy with standard object detection methods. Both our patch selection and target search approaches are seen to significantly outperform baseline strategies.

Whilst adversarial attack detection has received considerable attention, it remains a fundamentally challenging problem from two perspectives. First, while threat models can be well-defined, attacker strategies may still vary widely within those constraints. Therefore, detection should be considered as an open-set problem, standing in contrast to most current detection strategies. These methods take a closed-set view and train binary detectors, thus biasing detection toward attacks seen during detector training. Second, information is limited at test time and confounded by nuisance factors including the label and underlying content of the image. Many of the current high-performing techniques use training sets for dealing with some of these issues, but can be limited by the overall size and diversity of those sets during the detection step. We address these challenges via a novel strategy based on random subspace analysis. We present a technique that makes use of special properties of random projections, whereby we can characterize the behavior of clean and adversarial examples across a diverse set of subspaces. We then leverage the self-consistency (or inconsistency) of model activations to discern clean from adversarial examples. Performance evaluation demonstrates that our technique outperforms ($>0.92$ AUC) competing state of the art (SOTA) attack strategies, while remaining truly agnostic to the attack method itself. It also requires significantly less training data, composed only of clean examples, when compared to competing SOTA methods, which achieve only chance performance, when evaluated in a more rigorous testing scenario.

This paper considers attacks against machine learning algorithms used in remote sensing applications. The remote sensing domain presents a suite of challenges that are not fully addressed by current research focused on natural image data. In this paper we present a new study of adversarial examples in the context of satellite image classification problems. Using a recently curated data set and associated classifier, we provide a preliminary analysis of adversarial examples in settings where the targeted classifier is permitted multiple observations of the same location over time. While our experiments to date are purely digital, our problem setup incorporates a number of practical considerations that an attacker would need to take into account when mounting physical attacks.

We present a new dataset, Functional Map of the World (fMoW), which aims to inspire the development of machine learning models capable of predicting the functional purpose of buildings and land use from temporal sequences of satellite images and a rich set of metadata features. The metadata provided with each image enables reasoning about location, time, sun angles, physical sizes, and other features when making predictions about objects in the image. Our dataset consists of over 1 million images from over 200 countries. For each image, we provide at least one bounding box annotation containing one of 63 categories, including a false detection category. We present an analysis of the dataset along with baseline approaches that reason about metadata and temporal views. Our data, code, and pretrained models have been made publicly available.