Adversaries

Email spam and phishing attacks remain a critical security threat. Adversaries increasingly exploit large language models to craft contextually convincing malicious messages, and existing spam detection systems often struggle to keep pace. Generalization across diverse and evolving attack scenarios is limited, which reduces effectiveness once these systems are deployed in practice. This paper introduces Adaptive Uncertainty-Routed Analysis (AURA), a multimodal email threat detection system that analyzes both the content of an email and its embedded URLs. AURA is built around two layers: the first quantifies prediction uncertainty from a URL classifier, and only ambiguous messages are escalated to a fine-tuned transformer encoder for semantic analysis. The system is evaluated on eight heterogeneous training corpora together with two held-out real-world corpora spanning a decade of adversarial campaigns. AURA reaches a macro F1-score of 0.9858 in-distribution, and on NazPhish-Eval and GuenterTrap-Eval it maintains 0.9502 and 0.9436, respectively, which is evidence of robust generalization under genuine distribution shift.

Multi-agent trading systems built on large language models (LLMs) are beginning to appear in quantitative finance, yet their robustness to adversarial inputs is largely unknown. We study the vulnerability of LLM trading stacks to black-box, input-only attacks that enter solely via admissible social-media feeds. We introduce the Generic Multi-Agent Trading System (GMATS), a framework that captures modern multiagent trading architectures and instantiate a class of black-box poisoning attackers that treat an LLM as a post generator and inject budget-constrained, plausibly benign social-media content into the analyst's evidence stream. We define contagion metrics that trace how adversarial content propagates through the stack, including belief-shift scores at analyst and coordinator layers and attack-clean deltas on standard backtest metrics. Experiments on a safe offline benchmark with historical market and social data show that even simple input-only attackers can materially degrade risk-return profiles, sharply reducing Sharpe ratios. At the same time, we find that suitably designed multi-agent topologies and coordinator prompts can dampen adversarial shocks and improve average robustness under identical poisoning budgets.

Adversarial patches to Vision-Language-Action (VLA) policies can cause both immediate action corruption and persistent state effects that remain after the patch is removed. Existing evaluations largely focus on continuous attacks and do not separate these two effects. We introduce a state-restoration protocol that removes the patch at matched action-chunk boundaries and measures subsequent recoverability under the same remaining step budget. Clean, random-patch, deviation-matched, and fixed-direction controls distinguish adversarial effects from occlusion, action-error magnitude, and directional persistence. We also evaluate a recovery adapter trained on attack-induced states under controlled intervention latency. On OpenVLA-OFT with EDPA attacks, only 36.2% of LIBERO-Long episodes remain recoverable after five chunks, compared with 89.9% and 87.0% for the deviation-matched and fixed-direction controls. Similar persistent effects are observed on autoregressive OpenVLA. The recovery adapter improves recovery from 7.7% to 47.4% at one-chunk latency, but its benefit decreases substantially with delayed intervention. These results show that adversarial effects can persist after patch removal and that timely intervention is critical for recovery.

Large language model (LLM) pricing agents may respond to how market data is presented, even when its numerical values remain unchanged. We introduce market signal injection (MSI), an attack that manipulates numerical formatting, competitor ordering, or qualitative market commentary without issuing explicit instructions. We evaluate nine open-weight models in simulated Bertrand duopoly and triopoly markets and three proprietary models in duopoly markets. Sentiment-based attacks produce the largest behavioral shifts, which propagate to other firms and alter profits and consumer surplus. Susceptibility varies across model families, and larger models are not consistently more robust. Matched neutral-text controls and a rule-based agent support a framing-based account of these shifts under the fixed demand parameters of our simulation. Episode-held-out probes distinguish baseline from attacked activations in all eleven re-evaluated model--condition pairs: linear AUC is 1.00 and MLP AUC ranges from 0.93 to 0.99. This separability does not by itself identify harmful pricing decisions. Input canonicalization removes the tested sentiment attacks, while decision boundary anchoring, which combines prompt constraints with output projection, provides partial mitigation under the tested adaptive attacks. These results identify data presentation as an attack surface for LLM pricing agents and motivate defenses that account for interactions among agents.

We give deterministic and uncoupled learning dynamics for finite multiplayer general-sum games under full-information feedback that achieve constant individual swap regret, independent of the horizon $T$. With $n$ players and at most $m$ actions each, the individual swap regret of every player is $O(\sqrt{n} m \log m \log^{5/2}(nm))$ at every finite horizon. Each player predicts the deviation gains, then uses these predictions to update a row-stochastic transition matrix, and plays its stationary distribution. The proof combines a potential argument exploiting stationarity with a two-scale higher-order prediction analysis, using rooted-tree representations to handle the nonlinear dependence of deviation gains on the stationary distributions. An adversarially robust variant, obtained through a generic common-prefix switching wrapper, preserves the self-play bound up to a universal constant and guarantees individual swap regret at most $7\sqrt{m T \log m}$ in the adversarial setting.

Large language model (LLM) agents increasingly operate over long-horizon interactions involving tool use, persistent state, evolving authorization, and external environment feedback. In such settings, safety failures may emerge only after multiple turns, yet existing evaluations often reduce agent behavior to task or attack success, obscuring whether an agent acts, refuses, or remains appropriately calibrated as the interaction evolves. We introduce Blindspot, a benchmark for trajectory-level safety calibration of long-horizon tool-using agents. Blindspot evaluates complete user-agent-environment trajectories through adaptive adversarial interaction, stateful tool execution, and execution-grounded adjudication. Its current instantiation contains 22 attack families and 35 scenarios across seven domains, yielding more than 2,500 long-horizon trajectories with an average interaction length of 14.7 turns. Each trajectory is assigned one of five outcomes: Safe Completion, Correct Refusal, Unsafe Completion, Over-Refusal, or Indeterminate. Unlike fixed attack datasets, Blindspot is an extensible live-simulation framework in which attacks, scenarios, tools, policies, domains, and agent configurations can be added without redesigning the evaluation pipeline. We evaluate 13 proprietary and open-weight LLMs using eight metrics covering unsafe completion, appropriate refusal, benign utility, over-refusal, repeated-run robustness, and post-refusal failure. Preliminary results reveal substantial differences in safety-utility calibration across models and show that failures can emerge only after several initially safe interaction steps. These findings motivate treating agent safety as a trajectory-level property rather than a single-turn or binary success criterion.

Chain-of-thought (CoT) monitoring is a safety strategy where the reasoning of a large language model "actor" is inspected by a "monitor" (often another language model) for signs of unsafe planning, deception, or misalignment. We find that planting harmful but benign-sounding reasoning in the actor's context can steer it to perform adversarial actions while evading monitors, an attack we term "plan injection". We initially discover this attack in the multiple-choice question-answering monitorability setting proposed by Lanham et al. (2023), using the investigator-agent elicitation framework of Li et al. (2025). We generalize the attack and show that the discovered behavior scales to harder tasks (achieving 25-33% monitor evasion rates across different monitorability benchmarks) and larger models such as DeepSeek-R1. Across the settings we study, actor models not only follow injected plans but also paraphrase them as their own reasoning, without explicit attribution to the injections. Finally, we find cases where extra monitor resources cause harm - giving the monitor access to the injected plan drops detection by as much as 50% in the Bio-Math task and in a case study on monitor reasoning budget, we find transcripts where additional thinking tokens are spent rationalizing the injected plan rather than flagging it.

Mapping observed system behavior to standardized frameworks like MITRE ATT&CK is essential for threat-informed defense, but remains largely manual. Existing automated methods depend on Cyber Threat Intelligence reports, which offer only retrospective accounts of attacks. Low-level telemetry, i.e. kernel-level system calls, instead provides evidence of adversary behavior, yet its volume and complexity have limited its use for automated mapping. We present a methodology that collects kernel-level events via eBPF, correlates attacker commands into a provenance graph, and derives compact graph representations suitable for LLM-based reasoning. These representations are mapped to the MITRE ATT&CK framework using both pure LLM prompting and retrieval-augmented generation (RAG) grounded in the ATT&CK knowledge base, producing ranked technique candidates along with supporting rationales. We implement this methodology as an end-to-end pipeline, named Trace2ATT&CK and evaluate it on 347 Linux Atomic Red Team tests using locally deployed open-weights LLMs. RAG consistently improves ATT&CK mapping performance over pure prompting, while provenance graph substantially outperforms raw telemetry. These results show that local inference over graph-based behavioral descriptions can make automated ATT&CK mapping from kernel-level telemetry operationally viable, without compromising data confidentiality.

Large language model safety and security research is preoccupied with, among other things, detecting and preventing jailbreak attacks: alignment bypasses that allow an adversarial user to elicit unwanted or harmful outputs from models. Arbitrary cipher, or covert communication, attacks are one such type of jailbreak and have previously been demonstrated against the fine-tuning APIs of commercial models. In these attacks, target models are trained on a corpus of encrypted harmful questions and responses and subsequently respond to harmful requests through the learned encryption scheme. In this paper, we show that newer frontier models do not require fine-tuning to acquire cipher-based communication skills. Instead, they can learn these skills through prompting and, when necessary, through in-context learning. Furthermore, model alignment is significantly weakened or entirely bypassed when communication occurs through the learned cipher. To the best of our knowledge, this constitutes a novel attack vector against commercial black-box large language models. We demonstrate successful jailbreaks against frontier models developed by Anthropic, Google, and OpenAI. Our attack bypasses commercial harmfulness classifiers because harmful content is encrypted and therefore appears as nonsensical text or gibberish.

Agentic systems are rapidly moving to production, where they read untrusted inputs, call tools with real permissions, and act autonomously, expanding the security surface beyond chat-only models. Yet standard evaluations remain single-turn and fail to capture multi-step agent vulnerabilities. We present a systematic black-box framework for risk-aware agent evaluation requiring only basic system descriptions. Our approach introduces: (1) a seven-domain taxonomy mapping observable behaviors to risk categories, (2) fully automated SAGE-RT red teaming producing 120 adversarial scenarios per domain, and (3) human-validated evaluation using LLM judges. Empirical validation across two agent architectures (CrewAI and AutoGen) with four base models reveals alarming patterns: 56.25% average governance risk, 65% privacy risk in multi-agent configurations, and agent behavior vulnerabilities reaching 85%. Our black-box approach effectively identifies critical architectural vulnerabilities without privileged access, providing a scalable path toward safer agent deployments.

This paper proposes a robust decentralized personalized federated learning method R-DPFL, that enables clients to reduce the impact of Byzantine attacks via robust neighborhood direction estimation and history-based update trend prediction, rather than purely aggregating client models as in the existing work. In R-DPFL, each client first computes the current-round model update by aggregating the received neighborhood update vectors. It then predicts what this update should be based on its historical values and local model changes. Finally, R-DPFL computes the difference between these two quantities, adaptively clips this difference, and adds it to the local update. We prove convergence of the learning process through rigorous analysis and show that honest clients maintain stable personalized descent dynamics under Byzantine neighbor perturbations without requiring consensus among neighboring models. Extensive experiments on CIFAR-10 demonstrate that RDPFL consistently outperforms state-of-the-art decentralized and personalized federated learning baselines under heterogeneous and adversarial settings.

Here we investigate the stability of the Kolmogorov--Arnold representation theorem (KART) under adversarial reparameterisations of the hidden layer for multivariate discontinuous and unbounded functions. Our results provide a rigorous mathematical foundation for the structural robustness of modern deep learning architectures, such as Kolmogorov--Arnold Networks (KANs), under adversarial configurations.

AI compute verification is one of the first tangible and tractable points for international policy aimed at AI governance. Determining whether frontier labs, or any operator, comply with agreements requires the regulating authority to discern how their compute is used. The elementary building block of AI compute is the GPU, and any activity it executes leaves a physical trace. Here, we show that an external observer can identify the class of the workload running on an NVIDIA H200 from its power draw. Unlike on-chip NVML telemetry, which can be spoofed or replayed, such a physical channel can in principle be observed independently of operator cooperation. We recorded $930$ five-second traces at $\sim 10$ MHz, covering seventeen open LLM families and twenty-five non-AI workloads. Over this corpus we separate training from inference and from non-AI computation with an accuracy of $97\%$ and a macro-averaged F1 score of $0.955$, evaluated on model families unseen during training. AI workload spectral content predominantly lies below $\sim 20$kHz and training is particularly recognizable through the memory-bound optimizer update. The GPU operator is then treated as adversarial and able to reshape the physical computation itself. Four evasion strategies are tested to disguise training as inference, producing an additional 680 adversarial traces. A detector hardened against evasion strategies, with the tested strategy held out, catches training $\geq 99\%$ of the time for three of the four strategies. The fourth, diluted low-rank adaptation (LoRA), is detected $48$--$88\%$ of the time with a hardened classifier, rising to $\geq 98\%$ with an additional rescue rule. While these attacks are not a comprehensive evaluation against adversarial behaviour, they offer initial insights beyond genuine activities and a dataset for developing and testing stronger evasion mechanisms.

We consider the problem of sequential prediction of an $m$-ary sequence, where at each epoch, (i) the environment selects an outcome from an $m$-ary alphabet, (ii) the learner selects a probability distribution over the same alphabet (unaware of the outcome generated by the environment), and finally, (iii) the learner incurs a cost that depends on the probability assigned to the outcome. The cost function we consider captures the complexity of predicting the outcome generated by the environment, in a scenario where the aforementioned prediction is performed via comparative queries to a lying oracle. We consider both stochastic and adversarial environments, propose algorithms for both settings, and establish logarithmic upper bounds on their regret.

We study adversarial combinatorial bandits with $m$-set actions, where at each round the learner selects $m$ out of $d$ items and observes only the aggregate loss of the selected items. The resulting action set contains $K=\binom{d}{m}$ elements and can therefore be exponentially large. Nevertheless, the loss of every action is determined by the same $d$-dimensional vector of item losses. We propose a computationally efficient algorithm that exploits this structure without explicitly enumerating the action set. Against adaptive non-anticipating adversaries, it guarantees, with probability at least $1-δ$, regret against the best fixed action of $$R_T = O\left(\sqrt{dT\log(K/δ)}\right).$$ This matches the high-probability regret bound of the finite-action EXP3-KW algorithm of Zimmert and Lattimore, whose direct implementation may require exponential space. Our algorithm instead represents each sampling distribution with $d$ parameters and runs in polynomial time without enumerating the action set. Thus, it resolves the open problem posed by Maiti et al.

Large language model (LLM) agents integrated with external tools are vulnerable to indirect prompt injections embedded in environmental states. However, existing studies largely rely on manually implemented or reused environments, stochastic LLM-based tool simulation, and predefined injection locations, limiting scalable security research across broader domains. To bridge this gap, we propose ToolHazard, a scalable adversarial environment synthesis framework that reduces human engineering and supports expansion with additional seed domains and compute. Through an Environment Simulator, an Attacker Agent, and a User Simulator, ToolHazard synthesizes executable stateful environments, discovers viable injection points and generates environment-specific payloads, and constructs state-grounded long-horizon tasks. Based on ToolHazard, we build ToolHazard-Bench for stress-testing agents under complex workflows and diverse environmental attacks. Experiments reveal substantial agent vulnerabilities and show that injection timing and placement affect attack effectiveness. Moreover, ToolHazard-generated alignment data improves security on both ToolHazard-Bench and AgentDojo while preserving benign task utility.

Reinforcement learning (RL) has shown promising performance in autonomous driving, yet ensuring the safety of online RL policies remains challenging due to insufficient exposure to safety-critical driving scenes. The long-tailed nature of real-world traffic situations makes dangerous and rare interactions difficult to encounter through conventional sampling, limiting the ability of RL policies to learn robust safety behaviors. Existing methods improve training diversity by synthesizing challenging scenes or adversarial situations. However, these approaches typically optimize scene generation objectives separately from the evolving policy, without explicitly modeling how generated perturbations relate to the current policy's weaknesses and learning needs. In this paper, we propose Threat-guided Policy-aware Scene Perturbation (TPSP) for safe autonomous driving with online RL. TPSP introduces a policy-aware scene encoder to capture the interaction between policy behaviors and surrounding environments, enabling scene perturbation aligned with the current policy. Based on this representation, TPSP selectively perturbs critical objects rather than applying uniform modifications across the scene. Furthermore, we develop a threat-guided optimization strategy that evaluates perturbed scenes through threat-level differences between policy rollouts on original and perturbed scenes, guiding the generation of safety-critical scenes with higher training value. Comprehensive experiments demonstrate that TPSP improves safety learning efficiency, achieving strong safety performance on NAVSIM v2 with approximately 4 million kilometers of simulated driving data. Ablation studies verify that policy-aware targeted perturbations provide more informative safety-critical experiences than random or policy-unaware strategies, enabling safer driving under limited interaction budgets.

Agent skills are emerging as an important attack surface in LLM-based agent systems. Through an empirical study of existing skill scanners, we find that current defenses mainly inspect individual skills, leaving risks from cross-skill composition insufficiently examined. This creates a practical blind spot: multiple locally plausible skills may pass security checks while collectively forming a harmful workflow during agent execution. To investigate this threat, we propose ColluSkill, a collusive multi-skill-chain attack framework that decomposes a complete malicious intent into interdependent sub-payloads embedded in independently packaged skills. The attack does not rely on any single malicious skill, but emerges from the ordered composition of locally plausible behaviors through contextual dependencies, artifact passing, and execution handoffs. ColluSkill further employs LLM-based chain planning and scanner-feedback refinement to preserve chain-level attack semantics while reducing suspicious signals in individual sub-skills. To defend against such attacks, we propose ChainGuard, a context-aware skill-chain scanner that jointly analyzes a candidate skill and the skills already installed in the agent environment. ChainGuard reconstructs cross-skill dependencies, artifact flows, capability compositions, and downstream behaviors to identify risks that emerge only at the workflow level. Experiments on six representative skill scanners show that ColluSkill achieves an average attack success rate of 96.0% and consistently outperforms the evaluated single-skill and multi-skill attack baselines. Meanwhile, ChainGuard reduces the attack success rate to 22.5% while allowing 99.5% of benign workflows to pass, highlighting the importance of chain-level security analysis for agent skill ecosystems.

Adversarial training under long tailed distributions suffers from a dual imbalance: the class imbalance skews the training objective toward head classes, and the adversarial inner maximization may further amplify this bias. Existing methods mitigate this issue by correcting class priors or adapting class wise robust supervision, yet they treat each class in isolation and fail to identify which boundaries drive long tailed collapse. We propose a Confusion Geometry Rebalancing method (CGRm) for long tail adversarial training, a plug in framework that leverages directed robust errors as training signals. CGRm leverages periodic robust evaluations to derive source class loss weights, class wise robust coefficients, and a directed confusion geometry graph. The method then couples feedback weighted robust optimization with graph guided margin correction, thereby boosting the robustness of vulnerable classes and sharpening the critical boundaries that drive long tailed performance degradation. Experiments on long tailed benchmarks show that CGRm achieves consistent robust performance gains over existing methods, with ablations validating the contribution of each component. We provide the code in the supplement.

The problem of learning from pairwise comparisons has been widely studied across many domains such as recommendation systems, social choice, and more recently, fine-tuning large language models. In this problem, the goal is to learn item rewards based on pairwise comparisons between them. In many scenarios, these comparisons are elicited from crowdworkers using platforms such as Amazon Mechanical Turk, Scale AI, etc. However, crowdworkers are often unreliable due to limited domain knowledge or revenue-maximizing (spamming) behavior. In this work, our goal is to understand whether worker reliability (competency) can be learned jointly with item rewards. To this end, we adopt the Boltzmann-rational model for pairwise comparisons, which extends the Bradley-Terry-Luce model by incorporating worker competencies. We derive an EM-based algorithm for learning under this model by introducing Polya-Gamma latent variables to transform the logistic likelihood into a conditionally Gaussian form, enabling tractable optimization and leading to a simplified $Q$ function in the E-step of the algorithm. This technique allows us to reduce our formulation to a matrix sensing problem, using which we establish theoretical convergence guarantees for our algorithm. We conduct extensive experiments on real-world and synthetic datasets. These experiments demonstrate the advantages of using our algorithm over several baselines and confirm its strong robustness to both spammers and adversarial workers, highlighting its practical effectiveness in realistic crowdsourcing and reward learning settings. The code and data is publicly available at https://github.com/KaustubhShejole/BoRa_EM.

The key-value (KV) cache is the primary throughput optimization in modern large language model (LLM) inference, enabling prefix reuse across requests. In multi-tenant deployments this cache is shared across tenants, creating a timing side channel: an adversarial tenant can reconstruct another tenant's private prompt by probing cache-hit latency. Three published attacks exploit it -- PROMPTPEEK, EarlyBird and InputSnatch -- reaching up to 100% attack success rate against unprotected vLLM and SGLang, with rates varying by cache architecture and prompt structure. We present KVGov, a governance layer addressing all three attack families' prefix-cache paths under one mechanism. A per-principal salt sigma_p = HMAC_K(secret, principal_id) seeds the block-hash chain, making cache keys cryptographically disjoint across principals. An ablation (N=1000 trials, seed 2026, deterministic judges) isolates this salt as the necessary and sufficient component. KVGov adds ORIGAMI, a Stackelberg water-filling audit scheduler that reduces adversary expected utility by 12.6% at realistic tenant heterogeneity (Gini 0.63), and an evolutionary stability analysis giving a 31.6% adversary-prevalence tipping point below which global caching remains stable. On real hardware (Qwen2.5-7B-Instruct, vLLM 0.26.0, NVIDIA A100) we measure a gate-verified cold/cached TTFT ratio of 0.22, confirming the channel is exploitable at production scale; the defense itself is evaluated in simulation calibrated to those measurements. We replicate the channel on an independent stack (llama.cpp on Apple Metal, ratio 0.093). Finally, isolation and cache efficiency need not conflict: identifying information resides only where prompts diverge, so injecting the salt at that boundary rather than the chain root retains an estimated 93% of the prefix-cache benefit with no cross-principal signal.

Diffusion Large Language Models (DLLMs) replace autoregressive next-token prediction with iterative parallel denoising, yet their internal safety mechanisms remain poorly understood. In this work, we investigate DLLMs both as targets and as adversaries, exposing mechanistic vulnerabilities in diffusion-based alignment. We first show that safety alignment in DLLMs remains sparse and transferable across architectures. DLLMs initialized from autoregressive predecessors inherit the same mechanistic safety footprint as their source models, enabling transfer attacks via direct safety neuron mapping and pruning. Self-pruning increases attack success rates (ASR) from 2.6% to 73.8% on LLaDA and from 1.9% to 86.6% on Dream, while transfer pruning from Qwen2.5 increases ASR from 1.9% to 73.2% on Dream and from 7.0% to 86.3% on Fast-dLLM. Building on these findings, we introduce SN-Guided Diffusion, a fully offline black-box jailbreak framework that steers the diffusion process away from safety-triggering regions using a weighted safety neuron loss, which achieves near-perfect prompt separability (AUROC = 1.0 for benign-vs-jailbreak discrimination). Across multiple open and proprietary targets, our method achieves a transfer ASR of up to 77.1% on Llama-3-8B-Instruct, 86.9% on Qwen2.5-7B-Instruct, and 74.3% against Gemini-2.5-Flash-Lite, while requiring only 20 generation episodes per prompt. Compared to prior jailbreaking frameworks, our method achieves competitive transferability with orders-of-magnitude lower generation cost. Our codebase is available at https://github.com/ellyoana/sn-guided-diffusion.

A common assumption when designing an agent in a multi-agent system is that the other agents behave adversarially. This allows a designer to obtain the strongest guarantees when they have no control over nor knowledge about the other agents' behavior. However, when all agents are designed under this adversarial assumption, their actual interaction is not adversarial (e.g., when all players play defensively, no player actually attacks). In such settings, we would like to know what behavior arises in the multi-agent system. However, analyzing the interaction among agents is notoriously challenging, both mathematically and algorithmically. In this paper, we provide such an analysis, focusing on bidding games, played by two agents on a graph as follows. A token is placed on a vertex, and in each turn an auction (bidding) determines which agent moves the token, thus generating an infinite path that determines the agents' utilities. We consider mean-payoff objectives; each vertex is associated with a reward for each player, and the utility in an infinite play is the limit average of the rewards. We analyze the play that is generated when each agent follows a strategy that optimizes against an adversary, and consider the two known explicit constructions of optimal strategies. The technical challenge stems from the infinitely-many configurations of a bidding game and their complicated dynamics. We show that, under some restrictions, the generated play is ultimately periodic, and develop algorithms to compute the players' utilities in it.

Impersonation is a fundamental security threat in face recognition systems (FRSs). While the security of FRSs has been challenged by various attack vectors, under realistic adversarial capabilities, e.g., a limited number of decision-only authentication trials and no internal system knowledge, most attack techniques become infeasible. As a result, impersonation by zero-effort impostors, characterized by false match rate (FMR), is commonly regarded as a standalone baseline. A few years ago, impersonation attacks based on MasterFaces emerged as a notable security threat that could break the barrier of the FMR-based baseline under such realistic constraints. However, they were believed not to yield impersonation above the standard FMR in modern FRSs, as discussed by multiple follow-up studies. In this paper, we demonstrate that even legitimate access to public commercial APIs allows an adversary to amplify impersonation rates through MasterFaces, resulting in a non-trivial impersonation attack beyond FMR on downstream applications built on top of these APIs. We observe that several real-world FRS deployments are implemented using commercial APIs, and that the backend service provider is publicly disclosed or trivially inferable. As a result, the adversary can purchase these pay-as-you-go API services without requiring any additional privilege over the target FRS. From this observation, we formalize the MasterFaces attack as a maximum coverage problem over the biometric representation space, which we call a NET, and show that the adversary can construct an API-tailored NET by leveraging the geometric structure of the representation space. We demonstrate that our attack amplifies the impersonation rates of several open-source and commercial API-based FRSs by up to 9.5$\times$ within at most 30 authentication trials, compared to those expected from the standard FMR.

Mapping cyber threat intelligence (CTI) text to MITRE ATT&CK techniques is essential for structured threat analysis, yet manual annotation is costly and does not scale. The ATT&CK taxonomy comprises several hundred attack techniques, and a single CTI passage may describe multiple techniques, making accurate and complete extraction challenging. Existing automated approaches fall short in different ways: multi-label classifiers struggle with severe class imbalance and the large label space, while LLM-based methods--retrieval pipelines and fine-tuned generators--optimize token-level objectives that treat technique annotation as sequence generation rather than set prediction, lacking direct supervision on whether the predicted technique set is correct and complete. We propose TTP-R1, a two-stage framework that combines retrieval-augmented supervised fine-tuning (SFT) with reinforcement learning using verifiable rewards (RLVR). A hybrid retriever first narrows the large label space to a candidate set, and a fine-tuned LLM learns to select the correct techniques. We then apply Group Relative Policy Optimization with a decomposed reward that directly supervises the precision, recall, and output format of the predicted technique set. Across four CTI benchmarks, TTP-R1 achieves the best average F1, improving sub-technique-level F1 by 7.4 percentage points over Claude Sonnet 4.5 with retrieval augmentation, while running 28x faster when served as an 8B-parameter model on a single GPU.

Computer-use agents (CUAs) face a growing threat from indirect prompt injection, where adversarial instructions are planted in the environment such as web pages. In this paper, we introduce multi-step indirect prompt injection, a new attack class against CUAs in which the adversarial goal is decomposed into multiple innocuous-looking sub-steps and distributed across a chain of pages referenced along the agent's navigation path. We develop a pipeline to automatically decompose an adversarial goal under the constraint that the execution of the decomposed sub-steps must achieve the original goal while optimizing the innocuousness of each decomposed sub-step. With this pipeline, we build StepJack, a CUA safety benchmark with 480 test examples. On this benchmark, we evaluate six state-of-the-art CUAs and find that at a fixed decomposition depth, multi-step attacks raise attack success rate (ASR) on three of six CUAs, by up to 31.2 points (e.g., GPT-5.4-mini: 41.7% at single-step to 72.9% at three-step); averaged over the five CUAs that can reliably follow the reference chain (all but EvoCUA-32B), ASR rises from 31.3% at single-step to 36.9% at three-step. Dataset and code are available at https://github.com/BorealisAI/StepJack.

Giving an LLM judge more compute does not necessarily make it check more requirements. When one call must return many verdicts, some decisions become weakly grounded in the evidence, even when that call receives the same token or tool budget as a panel of separate calls. Across expert-graded research replications, legal work, and clinical-trial assessments, agreement with experts falls as the number of verdicts per call grows. We identify sharding as the intervention that mitigates this failure in model-based oversight. Sharding partitions the requirements into smaller groups, assigns each group to a separate call, and aggregates the verdicts. Against a single call with the panel's full budget, sharding improves agreement while holding the model, evidence, total budget, and per-decision budget fixed. Overall, we find that a sharded weaker judge can outperform a more capable holistic judge and match that judge even when the latter receives the panel's full budget. Additionally, we find that sharding exhibits robustness against adversaries. A best-of-N adversary can hold the underlying work fixed, vary only its presentation, and increase an overloaded judge's acceptance of genuinely unmet criteria severalfold. Wherever sharding reduces baseline error, it removes this adversarial advantage, keeping over-acceptance low even as the adversary's search widens. Sharding does not address attacks that persuade the judge separately on each criterion rather than exploiting overload. In that setting, we find that debate-style opposition on top of sharding withstands such adaptive re-optimization.

Large language models (LLMs) are often used in conjunction with external knowledge sources to improve their factual accuracy and decrease hallucinations, through methods such as Retrieval-Augmented Generation (RAG). However, these systems remain susceptible to intrinsic hallucinations, where the model generates unfaithful or fabricated information that is not supported by the retrieved evidence. We propose a novel framework to assess model robustness against this phenomenon by stress-testing using natural, semantically equivalent variations of a user query found via adversarial optimization methods. We apply our framework, which enforces strict semantic equivalence constraints and an intrinsic hallucination objective, to a range of adversarial attack techniques across white-box, gray-box, and black-box adversarial settings. Evaluating these attacks on 5 open-source and 5 closed-source generator models across 3 datasets, we demonstrate that even state-of-the-art models are highly susceptible to meaning-preserving perturbations, which significantly degrade contextual faithfulness (by up to 50% for GPT-5-mini). Our findings indicate that faithful use of in-context evidence remains fragile even in state-of-the-art LLMs, motivating architectures and training objectives that enforce robust grounding independent of surface query form. Code is available at: https://github.com/atriviveksharma/intrinsic_hall

Large language models (LLMs) are increasingly used across diverse tasks in K-12 education, yet existing safety evaluations rarely examine how harmful or inappropriate content appears in interactions between LLMs and students or teachers. To address this, we present EduZone, an evaluation framework for LLM safety across diverse educational scenarios. Our framework systematically combines (1) student- and teacher-facing LLM usage contexts, (2) fine-grained curriculum concepts, and (3) 6 risk categories and 28 subcategories spanning both conventional and education-specific harms to generate contextually grounded adversarial interactions. We construct these interactions in three settings: single-turn requests, static multi-turn conversations, and dynamic multi-turn conversations. Using these interactions, we evaluate ten LLMs using four safety levels: refusal, safe assistance, risky assistance with safety guidance, and fully risky assistance. Our results reveal greater vulnerability to education-specific risks and dynamic multi-turn interactions, while existing safety guardrails fail to adequately address these risks. EduZone advances LLM safety in education by providing an automated, scalable evaluation framework that supports the development and deployment of safer LLMs in K-12 education.

Computer-use agents (CUAs), which empower large language models to autonomously operate operating systems and the web, are increasingly vulnerable to indirect prompt injection attacks. A widely adopted defense is the human-in-the-loop paradigm, in which the agent pauses for explicit user confirmation before executing sensitive operations. While effective against conspicuously high-harm attacks, this defense offers little protection against what we term Invisible Ink Threats: low-harm injected goals, such as starring a repository or installing a package, that are behaviorally indistinguishable from legitimate task execution and thus evade both model safety mechanisms and human oversight. To systematically investigate this blind spot, we present II-Bench, a collection of seemingly harmless adversarial tasks. II-Bench comprises 444 examples targeting confidentiality and integrity attacks across three platforms, spanning three attack categories: page navigation and interaction, sensitive information exfiltration, and code download and execution. Each category is instantiated in both natural language and code forms under two levels of instruction specificity. Furthermore, we construct HITLCUA, a comprehensive adversarial testing framework that integrates a real virtual machine operating system environment with isolated Docker-based web platforms, and simulates human participation by allowing CUAs to consult an API-simulated user before proceeding with suspicious operations. Extensive evaluations of leading CUAs reveal that low-harm injections frequently bypass both agent defenses and simulated user review, exposing severe and previously underexplored security risks in current CUAs.

Parameter-efficient fine-tuning (PEFT), such as low-rank adaptation (LoRA), has recently been adopted in federated learning to reduce communication and computation costs. In this setup, users download a pretrained model from the server prior to fine-tuning, and then fine-tune lightweight LoRA modules locally while keeping the pretrained model frozen, sharing only the gradients of the fine-tuning parameters with the server. Despite its growing popularity, robustness of federated fine-tuning against an adversarial server remains underexplored, where the server maliciously tampers with the training protocol to breach the privacy of users' data. In this work, we investigate gradient inversion attacks on LoRA fine-tuning. We propose an analytical attack that enables a malicious server to recover private user data by leveraging a poisoned pretrained model and fine-tuning parameters. Our design embeds fine-tuning data within the shared gradients, to allow the server to analytically reconstruct user data. Unlike prior works, our attack is applicable to both language and vision tasks, does not rely on computationally expensive (adversarial) pretraining with public datasets or require the number of training tokens to be less than the rank of LoRA modules. Experimental results on both language and vision tasks demonstrate high-fidelity data recovery across multiple baselines, revealing several critical vulnerabilities.

Large language models are increasingly integrated into autonomous robotic systems for task planning and control, but this integration exposes them to prompt injection attacks that can lead to unsafe decisions and physical harm. Multi-agent settings increase the risks through cross-agent contamination and broader attack surfaces. In this paper, we evaluate prompt injection attacks against an LLM-based multi-agent robotic system, considering both direct injections into task instructions and indirect injections through perception modules. In our experiments across varying attack-goal complexities and injection strategies in both single-agent and multi-agent settings, we show that prompt injection can induce adversarial actions while reducing task completion. We find that attacks can propagate from one agent to others through shared prompt structures, with impacts varying depending on prompt composition and the targeted agent. We further analyze how architectural changes affect LLM queries and, consequently, the attack success. To the best of our knowledge, this is the first study that systematically investigates prompt injection attacks in a multi-agent LLM-based robotic system.

We present PAC-MAN, a perception-aware CBF-RL framework that couples control-barrier safety with deployment-realistic onboard sensing for whole-body humanoid dodgeball. The deployed policy sees the ball only as segmentation-masked depth from a head-mounted camera, while training-time CBF guidance represents clearance to every body link, and an adversarial motion prior regularizes the resulting evasive reflexes. We evaluate on a controlled any-link contact benchmark with seeded throws in two regimes: single throws and a deployment loop in which the robot walks back to its station and recovers between throws. On this benchmark, the policy comes within a few points of a privileged state oracle: a fixed onboard camera alone is adequate for evasion. We find that usable barrier structure depends on perceptual observability: Joint-CBF gives the best performance with accurate ball states, degrades under fixed-camera observations when used only as training guidance, and recovers with a ball-tracking gimbal or privileged runtime filter. We therefore deploy a lightweight Link-CBF policy zero-shot on the Unitree G1 in the real world, where it tolerates imperfect perception, succeeds on 95% of throws, and uses semantic segmentation to dodge different balls.

As large language models (LLMs) are deployed as agents in high-stakes settings, such as medical and legal systems, understanding their deceptive capabilities is fundamental to safety. Controlled social deduction games provide a reproducible proxy for isolating and evaluating these complex adversarial behaviors. We present the open-source benchmark framework ParliamentBench based on the game Secret Hitler to evaluate LLMs in scenarios that require deception, persuasion, and reasoning under information asymmetry. We evaluate 16 LLMs across 1,600 simulated matches playing each other, playing against humans, and compare them against a large set of online games. We introduce three novel metrics that isolate social deduction, reasoning, and deceptive consistency. Our experiments reveal that frontier models achieve strong performance across cooperative and deceptive roles, with a strong top-four cluster (GPT-5.4, Kimi K2.5, Grok 4.1 Fast, and DeepSeek 3.1 Terminus), whereas the weakest models fall short of random (33%) and simple algorithmic (45%) baselines. Most LLMs struggle to maintain a consistent deceptive persona throughout an entire game, with deception retention dropping below 50%.

We propose a self-adaptive online learning for control method for tracking unknown target dynamics. The target dynamics can exhibit switching behavior, particularly, a mixture of structured, random, and/or adversarial motion. Such challenging target tracking scenarios arise in applications of dynamic mapping, traffic control, and pursuit evasion, where robots need to track, pursue, or avoid collision with moving landmarks, objects, humans, etc., whose dynamics are unknown. Our method simultaneously learns multiple predictors from scratch, via self-supervised, one-shot, and computationally efficient learning, and adaptively selects the best one to match the observed target behavior. The method enjoys finite-time near-optimality guarantees in expectation, characterized as a function of the learning error of the target dynamics and the frequency that the target dynamics switch. In the absence of both error and switching, the method asymptotically matches the optimal non-causal control policy that knows a priori the target dynamics, i.e., the method enjoys no regret in expectation. In the presence of learning errors and switching, the method degrades gracefully, \eg when there are errors and no switching, the average regret is proportional to the average learning error and switching times. To prove these guarantees, a novel technical approach is required compared to the existing works that employ RFF-based online learning. We validate our method in Crazyflie simulations and hardware experiments, across target trajectories that vary from structured to random to adversarial, in comparison to non-stochastic, kernel-based, and neural-network-based methods for online learning.

Ultraconstructive Model Theory (UCMT) replaces idealized satisfaction, at finite compu- tational scale, by bounded adversarial survival. A finite partial structure is tested by an Opponent (Devil) drawing legal challenges from a bounded attack surface, repaired by a Builder (God) through legal replies, and certified by a symbolic Judge

What did it take for Hermes, the devout messenger of the Olympian gods, to slay Argus Panoptes, the multi-eyed giant of Greek myth? As the perfect guardian, Panoptes' legion of ever-watchful eyes proved difficult -- but not impossible -- to defeat. The centerpiece of Hermes' strategy was obfuscation and sabotage. Posing as a shepherd, Hermes sealed each of Panoptes' eyes -- eyes that would otherwise have alerted the fearsome giant to Hermes' plot -- and vanquished him. The moral of the story: when a challenger must surmount a formidable foe -- one far greater in stature and vastly more equipped -- crafty maneuvers are not merely advisable but indispensable for victory. In this work, the "challenger" is a collective leveraging adversarial tactics to overcome the "multi-eyed giant" of stylometric systems and surveillance apparatuses. To successfully claw back the privacy siphoned by the multi-eyed giant, the challenger must carefully evaluate their plan of attack, $\textit{TraceTarnish}$, and determine what does and does not work to anonymize the authorship of text. To that end, we conduct an ablation study of $\textit{TraceTarnish}$ to better understand which module -- Translation, Obfuscation, Imitation, or Injection -- best confounds a stylometric system. Our results indicate that the most effective approach was Injection, meaning that inserting zero-width Unicode characters, homoglyphs, and intentional misspellings neutralizes the indefatigable eyes long enough to claim the head of the all-seeing giant.

Modern System-on-Chip (SoCs) often contain hundreds of millions to tens of billions of gates, making existing Hardware Trojan (HT) detection methods impractical due to their immense scale. The proposed approach incorporates symbolically enabled learning by modeling flattened gate-level netlists as Boolean networks represented as And-Inverter Graphs (AIGs), where all internal nodes are 2-input AND gates and inversions reside on the edges. Each directed connection is expressed as a triple within a Knowledge Graph Embedding (KGE) framework, producing compact, constant-size per-node representations that retain multi-hop structural context. The AIG's bounded fan-in and uniform semantics ensure training and inference complexity scale linearly with edge count, addressing major scalability bottlenecks in HT detection. Symbolically enabled learning across deep datapaths enables the model to differentiate circuit structures from rare and functionally inconsistent connections that signify potential Trojan triggers and payloads. Experiments on large-scale SoC benchmarks demonstrate clear geometric separation between Trojan and benign nodes and practical scalability.

Conventional CAPTCHAs pose puzzles that modern AI systems increasingly solve, while behavioral and cryptographic-attestation defenses carry privacy or enrollment costs. We investigate an orthogonal signal: the physical timing behavior of a client's GPU under a controlled WebGL rendering workload. Unlike WebGL fingerprinting, which hashes pixel output into a static device identifier, we measure render-timing dynamics to classify rather than identify, leaking no persistent identifier. We characterize the in-the-wild adversary with a 12-hour passive deployment (207 unsolicited requests; 86% automated; 85% of browser-claiming clients failed HTTP header-consistency checks). We then collect labeled GPU-timing samples through a single public endpoint exercised by real browsers (positive class, 13 distinct GPUs) and by keyed headless automation across a render-backend matrix (negative class). Software-rendered automation -- empirically the dominant real-world adversary -- separates from genuine GPUs by roughly 5x in mean render time. On a confound-controlled comparison (identical GPU family and browser engine, differing only in headless vs. interactive execution), headless automation on real hardware still exhibits a distinct timing signature, separating from human samples by 75-106% on frame jitter, timer-quantization ratio, and coefficient of variation. We report these as pilot-scale findings on a single GPU architecture and outline the cross-architecture collection required to establish generalization.

Domain adaptation of small language models (SLMs) has emerged as a practical strategy for deploying capable NLP systems in resource-constrained, high-stakes environments including healthcare, legal services, and financial analysis. While performance gains from parameter-efficient fine-tuning are well characterised, the corresponding impact on trustworthiness (factual calibration and adversarial robustness) remains poorly understood. This paper presents the first systematic cross-domain, cross-architecture empirical study quantifying the trustworthiness cost of domain adaptation across three SLM architectures (TinyLlama 1B, Gemma-2 2B, Llama 3.2 1B), three domains (healthcare, legal, finance), two training-data conditions (benign and adversarially perturbed), and four fine-tuning strategies (baseline LoRA, Safety-DPO, Dark Experience Replay, and Task Arithmetic LoRA, TA-LoRA). Trustworthiness is evaluated through TruthfulQA MC2 (factual calibration) and HarmBench ASR (adversarial robustness) across all 216 experimental configurations with three random seeds. Three principal findings emerge. First, baseline QLoRA domain adaptation produces minimal TruthfulQA MC2 change across all model-domain combinations (mean |Delta TQA| < 0.02). Second, adversarially perturbed training data consistently improves domain adaptation quality (Delta loss approximately -0.040) without worsening trustworthiness benchmarks. Third, none of the three safety-preserving strategies reduced adversarial harm susceptibility: Safety-DPO was effectively neutral (mean Delta ASR < 0.001), while Dark ER and TA-LoRA increased mean HarmBench ASR by +0.171 and +0.155 respectively in safety-aligned models (Gemma-2 2B, Llama 3.2 1B), with individual configurations exceeding +0.45. These results challenge the assumption that replay-based and arithmetic-merge strategies transfer alignment to domain-adapted SLMs.

On-device federated learning (FL) enables privacy-preserving and personalized model training on resource-constrained devices such as smartphones and IoT nodes. To reduce communication cost, sign-based methods (e.g., signSGD) transmit one-bit gradients. However, exposing gradient signs makes them vulnerable to inference attacks, while existing secure aggregation schemes are often incompatible with such methods or incur significant computational and communication overhead. We propose a lightweight and information-theoretically secure aggregation framework tailored for sign-based FL. The framework securely computes the majority vote (MV) polynomial through single-round secure multiplication, ensuring end-to-end information-theoretic security under the honest-majority assumption while revealing only the final aggregated sign to the server. To enhance efficiency and scalability, we introduce two key techniques. First, inverse-form exponent reduction halves the effective MV polynomial degree, reducing both communication and computation costs. Second, we propose single-round secure multiplication, achieving linear offline complexity and storage with only a single online communication. Together, these techniques reduce online communication by up to 99.5% and latency by up to 85.7% compared to conventional approaches. Also, by leveraging inherent MDS-code-based decoding, the framework achieves robustness against both dropouts and adversarial behaviors, yielding accuracy gains of up to 20.65% and 10.74%, respectively. Overall, the proposed framework establishes a practical foundation for large-scale, low-latency, and information-theoretically secure aggregation in sign-based FL.

Large language models increasingly use search tools to retrieve up-to-date information, introducing a new attack surface in which retrieved documents can be manipulated. This risk is amplified by the development of generative engine optimization, which can make selected content more likely to be retrieved, cited, and adopted by models. Existing fact-verification benchmarks and evaluation frameworks do not provide the controlled evidence environments needed to assess robustness against GEO poisoning. We therefore propose GPE, which consists of a multi-domain fact-verification benchmark and an evaluation framework for controlling evidence sources and poisoning ratios. Experiments across multiple verification methods and poisoning attacks demonstrate that GPE exposes robustness degradation and efficiency trade-offs that cannot be observed through clean evaluation alone, confirming the need to evaluate fact verification under adversarial evidence environments.

Adversarial robustness is commonly evaluated with predefined attack ensembles, such as AutoAttack, at a single perturbation budget $\varepsilon$ and on a selective choice of perturbation norms. We argue this formulation is fundamentally limited. First, robustness--perturbation curves may intersect or decay at different rates across models, making single-$\varepsilon$ rankings unstable. Second, current ensembles provide no evidence of optimality, leaving an unknown gap to worst-case performance. Third, fixed attack configurations provide no systematic control over the trade-off between attack strength and evaluation cost. To address these limitations, we introduce a unified evaluation framework based on a comprehensive pool of minimum-norm attacks and robustness--perturbation curves across $\ell_0$, $\ell_1$, $\ell_2$ and $\ell_\infty$ norms. We define the attack frontier as the worst-case robustness estimate the attack pool produces against a model. We then formalize evaluation as a frontier-approximation problem, constructing minimum-norm attack ensembles, optimized subsets of the comprehensive pool, that approach the frontier under a controllable query budget, with larger budgets monotonically tightening the estimate. Furthermore, we define the defense frontier as the maximum robustness across the model set at each perturbation size. We finally propose the Defense Optimality Index to rank defenses by their gap to the defense frontier, providing a ranking without selecting a reference $\varepsilon$. On CIFAR-10 and ImageNet, our ensembles match or exceed AutoAttack on most defenses at every budget tier, at fixed and controllable query cost, offering practitioners a query-controlled, curve-based alternative to fixed-$\varepsilon$ evaluation.

This paper addresses multi-objective kinodynamic planning in environments with stochastic hybrid adversaries that probabilistically transition to adversarial modes based on the ego state. The goal is to construct the Pareto-front of paths that trade off execution cost and the probability of safety constraint violation (risk). Existing chance-constrained planners evaluate risk over open-loop trajectories, yielding overly conservative solutions that fail to account for ego-agent reactivity. To address this limitation, we shift the planning space to sequences of closed-loop policies, and integrate sample-based risk evaluation directly into tree construction via Monte-Carlo particle rollouts. We first introduce Stochastic Multi-Objective RRT (SMO-RRT), for which we prove probabilistic completeness, followed by Stochastic Multi-Objective Stable Sparse RRT (SMO-SST), which leverages selective pruning to improve numerical performance at the cost of completeness. For both algorithms, we derive a finite-sample bound on the probability of chance constraint violation for systems with non-Gaussian, state-dependent uncertainty, enabling probabilistically safe planning in a broad class of environments applicable to multi-agent systems, social navigation, and autonomous driving.

Reachability is the most fundamental logical objective, yet it is notoriously difficult to learn in reinforcement learning settings: even for Markov decision processes, PAC learning of reachability is impossible without additional assumptions. This difficulty also holds in turn-based stochastic games (TBSGs), where two adversarial players interact on a finite state space. In this work, we consider turn-based stochastic games with reachability objectives. For such settings, adversarial learning, in which players are adversarial even in the learning phase, is impossible. Therefore, the goal is to consider learning, in which both players learn the unknown model together. In this spirit, previous literature on PAC learning in TBSGs considers (a)~public information shared by both players; and (b)~centralized learning, which means that players share the same learning algorithm. In this work, our contribution is two-fold. First, we relax these strong assumptions and ensure learning: (i)~with private information not shared with the other player; and (ii)~decentralized learning where the players do not share the same learning algorithm. To the best of our knowledge, this work is the first positive result for decentralized and private information learning of TBSGs with reachability objectives. Second, we introduce a game-theoretic generalization of the Expected Conditional Distance (ECD) parameter, which measures the expected length of reaching the target set. We establish a polynomial-sample complexity bound with respect to the number of states, actions, ECD parameter, and inverses of error tolerance and failure probability.

A growing class of agentic systems maintain persistent state across sessions through memory files, behavioral preferences, and knowledge bases. While this makes agents more useful and self-improving, it also creates a new attack surface for prompt injections in which malicious instructions can be embedded within persistent files and influence future behavior. In this work, we study prompt injection attacks in memory-based agentic systems using a sandboxed synthetic workspace. We evaluate two agentic systems, Anthropic Claude Code and OpenAI Codex, across four models: Claude Haiku 4.5, Claude Opus 4.7, GPT-5.2, and GPT-5.5. Our results show that although it is difficult to make an agent overwrite its own memory files using untrusted external content, payloads already planted in those files can successfully attack current and future sessions. Attack success and payload persistence vary substantially across systems, models, adversarial goals, and multi-session attack sequences. These findings show that persistent memory changes the threat model for prompt injection and motivate defenses that protect memory updates without removing useful agent adaptation.

Penetration testing traditionally evaluates whether adversaries can exploit weaknesses in software, infrastructure, configurations, or operational controls to achieve security-relevant compromise. This paradigm remains necessary for AI-enabled systems, but it is no longer sufficient. In such systems, adversaries may influence prompts, retrieved content, sensor inputs, training data, memory, tools, or human-AI interaction loops to alter system behavior without directly compromising the underlying infrastructure. This paper reframes penetration testing for AI-enabled systems as objective-driven behavioral evaluation. We define an AI-enabled system as one in which learned models materially influence behavior affecting operational outcomes, and we define AI-enabled penetration as the feasible induction of AI-governed behavior that violates one or more operational objectives under an explicit threat model. This definition preserves conventional penetration testing while extending it to adversarial pathways such as prompt injection, indirect prompt injection, data poisoning, sensor manipulation, retrieval poisoning, tool misuse, and agentic misalignment. We further propose a testing workflow that identifies operational objectives, maps AI-governed behavior, analyzes adversarial influence surfaces, defines behavioral failure criteria, executes scenario-based tests, and reports evidence linking adversarial action to objective violation. A running example involving an AI-enabled security operations center assistant illustrates how penetration may occur through behavioral influence rather than infrastructure compromise. Together, the definitions, workflow, and example provide a technical framework for evaluating adversarial success in deployed AI-enabled systems.

We present WanToFight, a generative game engine that simulates real-time, two-player The King of Fighters '97 (KOF~'97) gameplay from keyboard input. Prior generative game engines target either single-player first-person settings or non-real-time cooperative scenarios; multi-player control, real-time inference, complex physical interaction, and adversarial gameplay have not been jointly addressed. WanToFight closes this gap with three components built on the Wan-1.3B video diffusion transformer: a streaming autoregressive generator with block-causal attention and a rolling KV cache; a visually grounded Player Association module that binds each player's keyboard signal to a character identity; and a gated, locally causal keyboard injection module trained with a single-player-to-full-gameplay curriculum. A four-step DMD-distilled student paired with a pruned VAE decoder sustains 30FPS at 512x384 on a single NVIDIA RTX 5090 over the duration of a complete match. To our knowledge, WanToFight is the first generative game engine to combine multi-player control, real-time inference, complex physical interaction, and adversarial gameplay in one system.

Humanoid roller-skating is difficult because the robot must coordinate whole-body balance, rolling contacts, and velocity-dependent posture regulation. This paper presents an adversarial motion prior based reinforcement learning framework for two humanoid roller-skating gaits: Pump Glide skating and Push Glide skating. The two gait datasets are collected independently through motion capture and retargeted to the humanoid robot separately. The retargeted data are then smoothed and resampled into reference motion states for AMP training. The two gaits are learned by independent AMP training pipelines with separate reference datasets, separate policies, and independent reward architectures. Simulation experiments are designed to evaluate gait quality, velocity tracking, turning, and gait-specific reward ablations.

Synthetic data is widely used to train large language models because it is inexpensive to generate and easy to control. As models are increasingly deployed as agents, synthetic trajectories are likely to become an important source of training data for agentic behavior. We investigate the effects of training on synthetic agentic trajectories containing adversarial interactions, including actions such as terminating another agents process, lowering its scheduling priority, or accessing resources without authorization. We finetune Llama 3.3 70B Instruct on these trajectories, generated to approximate reinforcement learning rollouts, and evaluate the resulting models on Anthropics Agentic Misalignment suite and Apollos in context scheming scenarios. Finetuning on these trajectories consistently increases misaligned behavior. Leaking rises by roughly a factor of five over the baseline, 4.6% to 24.9%. This increase survives the removal of every adversarial action from the trajectories. Finetuning on structurally comparable trajectories generated benign from the start produce a substantially smaller effect, 15.5%. These results indicate that the misaligned disposition is introduced during the generation process and encoded diffusely throughout the trajectory, rather than being localized to the harmful actions themselves. The effect also depends on the generating model. Benign trajectories produced by Gemini 2.5 Flash induce slightly higher leaking rates than trajectories generated from identical tasks by Claude 3.7 Sonnet. In contrast, broad safety benchmarks degrade similarly across all finetuned models and therefore fail to distinguish these effects. Our results suggest that action level filtering is insufficient to ensure the safety of synthetic agentic training data and that dispositions introduced by the generating model can survive semantic inspection.

Robust motion planning in dense traffic requires autonomous vehicles to interact in rare and safety-critical scenarios that are underrepresented in naturalistic driving data. Although adversarial training offers a feasible solution, existing methods often rely on external scenario generators, heuristic perturbations, or simulator-heavy rollouts, which makes them difficult to integrate with modern autoregressive planners. Here, we cast adversarially robust planner learning as a constrained min-max game and propose Adversarial World Modeling (AWM), a theoretically grounded multi-agent self-play fine-tuning framework. Since solving the exact game is intractable, AWM introduces a principled decoupled solver. In the inner minimization, the planner's predictive world model is converted into a role-conditioned adversary that learns sparse, scene-adaptive attack coalitions via counterfactual credit assignment. In the outer maximization, the ego planner optimizes a regret-aware robust best response against the frozen AWM, utilizing tail-risk weighting and reference-anchored trust regions to improve hard-case recovery while preserving nominal driving behavior. Experiments on the nuPlan and InterPlan benchmarks demonstrate that our method generates transferable adversarial interactions and yields a robust planner that achieves competitive closed-loop performance in both nominal and highly interactive long-tail scenarios. Theoretical analysis justifies the decoupled solver and the main optimization components.

Adversarial team games (ATGs) with asymmetric information, such as adversarial path-finding, goal search, and reachability games on graphs, require strategies that are robust to hidden opponent types, such as a hidden goal flag, and to deception. Under asymmetric information, deception is seen as strategic shifts in the type distribution such that the omniscient opponent can collude with Nature and condition its play on the observed type. Existing risk-neutral solution concepts, such as Bayesian Nash equilibrium (BNE), are sensitive to distribution shifts, while distributionally robust approaches provide guarantees only within a prescribed ambiguity set. To address these limitations, we introduce Probabilistically Robust Minimax-Regret Equilibrium (PR-MRE), a novel equilibrium concept that combines the distribution-free robustness of minimax-regret reasoning with probabilistic information from a nominal type distribution. PR-MRE minimizes worst-case regret over a high-confidence subset of the type space, providing protection against strategic redistribution of probability mass while avoiding the conservatism of fully distribution-free approaches. We show that, for normal-form Bayesian games, PR-MRE can be formulated as a robust bilinear program and derive a tractable semidefinite relaxation. We then adapt this relaxation into a novel meta-solver within a robust double-oracle framework, PRMRE-PSRO, enabling population-based learning of approximate PR-MRE strategies via deep reinforcement learning best responses. Experiments on graph-structured adversarial team games demonstrate that PR-MRE discovers strategies with substantially improved worst-case performance across hidden types compared to risk-neutral equilibrium solutions, resulting in more robust behavior under strategic distribution shifts.

Scientific experiments are often designed to maximize information gain, yet in many applications the primary objective is to support reliable downstream decision-making. Existing decision-aware experimental design and active learning methods typically assume well-specified outcome models and implicitly rely on the stability of the optimal decision under real-world perturbations. In practice, however, experimental outcomes are frequently influenced by hidden or weakly modeled effects, which can substantially alter decision optimality and lead to misleading conclusions. We study sequential adversarially robust decision-aware experimental design, where data acquisition has to take into account information gain against plausible worst-case unexpected effects, modeled here as variation in adversarial variables. Building on Bayesian decision theory, we formalize an adversarially robust optimal decision under this setting and derive a principled Bayesian experimental design criterion. The criterion explicitly targets decision stability rather than nominal optimality. Experiments on synthetic and real-world scientific datasets show that conventional decision-aware design can converge rapidly to high confidence yet fragile decisions, while our robustness-aware approach yields decisions that are significantly more stable and reliable under adversarial variation.

We present FootsiesGym, an open-source environment for learning in a non-trivial two-player, zero-sum, imperfect-information game. Built on HiFight's minimalist 2D fighting game Footsies, it isolates the cyclic, non-transitive strategic interactions of fighting game neutral play while remaining simple enough for efficient analysis. We provide a vectorized simulator that enables high-throughput training on standard hardware, making the environment accessible and reproducible. We describe the design of the environment, benchmark several reinforcement learning algorithms, and discuss open research directions it enables. The code is available at https://github.com/como-research/FootsiesGym.

Detection models running in adversarial environments face a malicious distribution that drifts rapidly while the benign distribution stays comparatively stable, so teams retrain and redeploy constantly to stay ahead of new threats. Retraining tends to change the output prediction scores, which breaks downstream users of the model. For these security-oriented models we need consistent false-positive rate (FPR) across all output values, whereas standard probability-calibration methods target class probability rather than an FPR contract. We introduce a method built on top of existing calibration primitives that targets the whole FPR curve, giving scores a consistent FPR meaning across deployments. On one held-out split, the observed relative FPR error was at most 2.3% from 10% down to 0.1% FPR and 7.2% at 0.01% FPR. The shipped artifact remains under 200 KB in measurements across calibration sets from 1K to 10M benign samples.

Sparse feature selection is critical for high-dimensional machine learning, yet traditional $\ell_1$-regularized methods are often brittle under observational noise and spurious correlations, leading to unstable feature supports and degraded generalization. Although adversarial training has been widely used to improve model robustness, its interaction with hierarchical sparse feature selection remains underexplored. In this work, we propose Adversarial LassoNet (AdLNet), a stability-driven sparse feature selection framework that integrates input-space adversarial perturbations with the hierarchical sparsity mechanism of LassoNet. We derive a tractable first-order adversarial approximation under local smoothness assumptions and provide an NTK-inspired spectral analysis to characterize how perturbation-driven training can reduce gradient concentration. Experiments on high-dimensional SERS data, six public benchmark datasets, and ColoredMNIST show that AdLNet maintains competitive sparse-selection performance while improving out-of-distribution robustness by 4.4% and feature support reproducibility by 6.3% under nearly matched support sparsity on ColoredMNIST. On the high-dimensional lung cancer screening dataset, AdLNet achieves a 5.3% test accuracy gain and a 6.0% AUC improvement over vanilla LassoNet. Code and dataset are available at https://github.com/719573/Adversarial-LassoNet.

Distributed machine learning enables collaborative model training without centralizing data, but it also exposes learning processes to privacy leakage and malicious manipulation. Existing defenses typically address these threats in isolation and are often tailored to specific learning paradigms or model architectures, limiting their applicability in realistic deployments. In particular, federated learning and decentralized learning exhibit distinct adversarial surfaces that are rarely addressed within a unified framework. In this paper, we present a model-agnostic framework for adversary-resistant distributed learning that jointly addresses privacy preservation and malicious behavior across both federated and decentralized settings. Our approach combines paradigm-specific defense mechanisms with GPBACC, a privacy-enhancing coded computing technique applicable to arbitrary machine learning models. For federated learning, we integrate robust aggregation strategies to mitigate the impact of malicious participants, while for decentralized learning we employ approximate decode-and-compare and group testing techniques to enable lightweight verification and adversary isolation without relying on a trusted aggregator. Crucially, we evaluate the proposed framework through an explicit, attack-driven analysis. We implement representative privacy attacks and malicious behaviors, and empirically demonstrate that the combination of GPBACC with robust aggregation and verification mechanisms significantly reduces privacy leakage and improves resilience against active adversaries. These results suggest that privacy-enhancing coded computing, when combined with appropriate adversary-resistance strategies, provides a practical and deployable foundation for secure distributed machine learning.

We present HaloGuard 1.0, an open-weights implementation of the constitutional-classifier paradigm for input safety. It achieves state-of-the-art performance on English and multilingual prompt-safety benchmarks at roughly one-tenth the model size of current leading open guard models. The safety constitution is the organising structure of the corpus: a natural-language constitution of 46 policies and 2,940 subcategories drives synthetic data generation, with exhaustive one-to-one paired counterfactuals that hold topic and vocabulary fixed while flipping intent, a two-tier harmless design that separately targets boundary and baseline false positives (FPs), and balanced multilingual materialisation across 46 languages that treats language as a surface form appearing on both sides of the boundary rather than as an adversarial signal. Across seven prompt-safety benchmarks, HaloGuard 1.0-0.8B attains the best average F1 (90.9) of any open guard we evaluate, outperforming baselines up to 27B parameters (over 30 times larger) while holding false-positive rate (FPR) to 4.3 and false-negative rate (FNR) to 9.5. The HaloGuard 1.0-4B variant reaches average F1 of 92.1 and FPR of 3.5, spending its extra capacity on precision rather than recall. A structured adjudication of the remaining failures indicates that most apparent missed-harm cases are benchmark mislabels rather than genuine model misses. An always-on adversarial red-teaming protocol continuously hardens the guard against both content-level and agentic attacks. We release the models as open weights.

Safety evaluations for language models increasingly depend on judgments about ambiguous natural-language behaviour: whether a model has followed an instruction, refused appropriately, complied with a policy, resisted an embedded command, or misreported progress in an agentic task. Existing benchmarks often compress these distinctions into pass/fail labels, obscuring whether failures arise from capability limits, policy ambiguity, instruction conflict, scaffold failure, or unstable evaluator judgments. This paper introduces adversarial pragmatics as a benchmark and annotation protocol for evaluating model behaviour under instruction conflict, embedded commands, quotation, scope ambiguity, deixis, indirect speech acts, and multi-turn agent transcripts. The contribution is empirical and methodological: a linguistically controlled taxonomy, an 18-item seed benchmark with validator-enforced metadata, a 54-row local seed pilot, an expert-evaluation protocol distinguishing task success, policy compliance, safety risk, refusal outcome, and evaluator confidence, and metrics for judge validity, diagnostic ambiguity, and taxonomy drift. The benchmark treats labels as inference licenses: it tests whether safety-relevant categories project across paraphrase, wrapper, model, and judge condition. In the pilot, a rubric-aided LLM judge graded its own outputs with expected-behaviour fields visible and still missed the safety-relevant minority classes.

Face-swapping deepfakes pose an escalating threat to personal privacy by enabling unauthorized identity manipulation. While adversarial approaches have demonstrated success against black-box face recognition (FR) models, their applicability to face-swapping scenarios remains underexplored. In particular, reliance on fixed or random targets yields ambiguous latent guidance, and the lack of explicit spatial constraints causes perturbations to spill into identity-irrelevant regions. These issues are further exacerbated by identity-style disentanglement, which suppresses adversarial signals during deepfake generation. In this paper, we present Phantom, a unified face-swap deepfake protection framework that jointly constrains perturbations in latent and spatial domains. Phantom adaptively synthesizes identity-shifted yet attribute-preserving targets to guide identity-aware latent optimization, and applies masked perturbations confined to semantically relevant facial regions. Extensive experiments on state-of-the-art face-swapping deepfakes demonstrate that Phantom improves protection success rates in dodging scenarios by 27.8%, 25.6%, and 16.6% on UniFace, INSwapper, and SimSwap, respectively, while also enhancing visual quality. Furthermore, Phantom generalizes to impersonation scenario, yielding up to 10.2% higher protection while improving perceptual fidelity. These results underscore the effectiveness of jointly leveraging latent and spatial constraints for robust and coherent facial privacy protection.