Papers

Backdoor attacks pose severe security threats to large language models (LLMs), where a model behaves normally under benign inputs but produces malicious outputs when a hidden trigger appears. Existing backdoor removal methods typically assume prior knowledge of triggers, access to a clean reference model, or rely on aggressive finetuning configurations, and are often limited to classification tasks. However, such assumptions fall apart in real-world instruction-tuned LLM settings. In this work, we propose a new framework for purifying instruction-tuned LLM without any prior trigger knowledge or clean references. Through systematic sanity checks, we find that backdoor associations are redundantly encoded across MLP layers, while attention modules primarily amplify trigger signals without establishing the behavior. Leveraging this insight, we shift the focus from isolating specific backdoor triggers to cutting off the trigger-behavior associations, and design an immunization-inspired elimination approach: by constructing multiple synthetic backdoored variants of the given suspicious model, each trained with different malicious trigger-behavior pairs, and contrasting them with their clean counterparts. The recurring modifications across variants reveal a shared "backdoor signature"-analogous to antigens in a virus. Guided by this signature, we neutralize highly suspicious components in LLM and apply lightweight finetuning to restore its fluency, producing purified models that withstand diverse backdoor attacks and threat models while preserving generative capability.

We present a technical case study on the Privacy-Enhancing Technologies (PETs) for Public Health Challenge, a collaborative effort to safely leverage sensitive private sector data for social impact, specifically pandemic management. The project utilized Differential Privacy (DP) to create realistic, privacy-preserved synthetic financial transaction data, which was then combined with public health and mobility datasets. This approach successfully addressed the critical hurdle of sharing sensitive financial information for research and policy. The analysis demonstrated that this synthetic, DP-protected data possesses significant spatial-temporal and predictive power for public health. Key outcomes include the development of six reusable tools and frameworks supporting diagnostic nowcasting (e.g., Hotspot Detection, Pandemic Adherence Monitoring) and predictive forecasting (e.g., Mobility Analysis, Contact Matrix Estimation) for epidemiological decision-making. The study provides best practices for advancing data sharing in a privacy-compliant manner.

Diffusion-based image-to-video (I2V) models increasingly exhibit world-model-like properties by implicitly capturing temporal dynamics. However, existing studies have mainly focused on visual quality and controllability, and the robustness of the state transition learned by the model remains understudied. To fill this gap, we are the first to analyze the vulnerability of I2V models, find that temporal control mechanisms constitute a new attack surface, and reveal the challenge of modeling them uniformly under different attack settings. Based on this, we propose a trajectory-control attack, called CtrlAttack, to interfere with state evolution during the generation process. Specifically, we represent the perturbation as a low-dimensional velocity field and construct a continuous displacement field via temporal integration, thereby affecting the model's state transitions while maintaining temporal consistency; meanwhile, we map the perturbation to the observation space, making the method applicable to both white-box and black-box attack settings. Experimental results show that even under low-dimensional and strongly regularized perturbation constraints, our method can still significantly disrupt temporal consistency by increasing the attack success rate (ASR) to over 90% in the white-box setting and over 80% in the black-box setting, while keeping the variation of the FID and FVD within 6 and 130, respectively, thus revealing the potential security risk of I2V models at the level of state dynamics.

Prompt injection remains one of the most practical attack vectors against LLM-integrated applications. We replicate the Microsoft LLMail-Inject benchmark (Greshake et al., 2024) against current generation models running inside OpenClaw, an open source multitool agent platform. Our proposed defense combines two mechanisms: agent isolation, implemented as a privilege separated two-agent pipeline with tool partitioning, and JSON formatting, which produces structured output that strips persuasive framing before the action agent processes it. We run four experiments on the same 649 attacks that succeeded against our single-agent baseline. The full pipeline achieves 0 percent attack success rate (ASR) on the evaluated benchmark. Agent isolation alone achieves 0.31 percent ASR, approximately 323 times lower than the baseline. JSON formatting alone achieves 14.18 percent ASR, about 7.1 times lower. Our ablation study confirms that agent isolation is the dominant mechanism. JSON formatting provides additional hardening but is not sufficient on its own. The defense is structural: the action agent never receives raw injection content regardless of model behavior on any individual input.

Prior approaches for membership privacy preservation usually update or retrain all weights in neural networks, which is costly and can lead to unnecessary utility loss or even more serious misalignment in predictions between training data and non-training data. In this work, we observed three insights: i) privacy vulnerability exists in a very small fraction of weights; ii) however, most of those weights also critically impact utility performance; iii) the importance of weights stems from their locations rather than their values. According to these insights, to preserve privacy, we score critical weights, and instead of discarding those neurons, we rewind only the weights for fine-tuning. We show that, through extensive experiments, this mechanism exhibits outperforming resilience in most cases against Membership Inference Attacks while maintaining utility.

Existing methods for verifying access control policies require the policy to be complete and fully determined before verification can proceed, but in practice policies are developed iteratively, composed from independently maintained components, and extended as organisational structures evolve. We introduce robust property verification: the problem of determining what a policy's structure commits it to regardless of how pending decisions are resolved and regardless of subsequent extension. We define a support judgment $\Vdash_{P}\phi$ stating that policy $P$ has robust property $\phi$, with connectives for implication, conjunction, disjunction, and negation, prove that it is compositional (verified properties persist under policy extension by a monotonicity theorem), and show that despite quantifying universally over all possible policy extensions the judgment reduces to proof search in a second-order logic programming language. Soundness and completeness of this reduction are established, yielding a finitary and executable verification procedure for robust security properties.

OpenClaw-like agents offer substantial productivity benefits, yet they are insecure by default because they combine untrusted inputs, autonomous action, extensibility, and privileged system access within a single execution loop. We use OpenClaw as an exemplar of a broader class of agents that interact with interfaces, manipulate files, invoke tools, and install extensions in real operating environments. Consequently, their security should be treated as a software engineering problem rather than as a product-specific concern. To address these architectural vulnerabilities, we propose a blueprint for defensible design. We present a risk taxonomy, secure engineering principles, and a practical research agenda to institutionalize safety in agent construction. Our goal is to transition the community focus from isolated vulnerability patching toward systematic defensive engineering and robust deployment practices.

Diffusion models enable high-fidelity image editing but can also be misused for unauthorized style imitation and harmful content generation. To mitigate these risks, proactive image protection methods embed small, often imperceptible adversarial perturbations into images before sharing to disrupt downstream editing or fine-tuning. However, in realistic post-release scenarios, content owners cannot control downstream processing pipelines, and protections optimized for a surrogate model may fail when attackers use mismatched diffusion pipelines. Existing purification methods can weaken protections but often sacrifice image quality and rarely examine architectural mismatch. We introduce a unified post-release purification framework to evaluate protection survivability under model mismatch. We propose two practical purifiers: VAE-Trans, which corrects protected images via latent-space projection, and EditorClean, which performs instruction-guided reconstruction with a Diffusion Transformer to exploit architectural heterogeneity. Both operate without access to protected images or defense internals. Across 2,100 editing tasks and six representative protection methods, EditorClean consistently restores editability. Compared to protected inputs, it improves PSNR by 3-6 dB and reduces FID by 50-70 percent on downstream edits, while outperforming prior purification baselines by about 2 dB PSNR and 30 percent lower FID. Our results reveal a purify-once, edit-freely failure mode: once purification succeeds, the protective signal is largely removed, enabling unrestricted editing. This highlights the need to evaluate protections under model mismatch and design defenses robust to heterogeneous attackers.

Prompt injection poses serious security risks to real-world LLM applications, particularly autonomous agents. Although many defenses have been proposed, their robustness against adaptive attacks remains insufficiently evaluated, potentially creating a false sense of security. In this work, we propose PISmith, a reinforcement learning (RL)-based red-teaming framework that systematically assesses existing prompt-injection defenses by training an attack LLM to optimize injected prompts in a practical black-box setting, where the attacker can only query the defended LLM and observe its outputs. We find that directly applying standard GRPO to attack strong defenses leads to sub-optimal performance due to extreme reward sparsity – most generated injected prompts are blocked by the defense, causing the policy's entropy to collapse before discovering effective attack strategies, while the rare successes cannot be learned effectively. In response, we introduce adaptive entropy regularization and dynamic advantage weighting to sustain exploration and amplify learning from scarce successes. Extensive evaluation on 13 benchmarks demonstrates that state-of-the-art prompt injection defenses remain vulnerable to adaptive attacks. We also compare PISmith with 7 baselines across static, search-based, and RL-based attack categories, showing that PISmith consistently achieves the highest attack success rates. Furthermore, PISmith achieves strong performance in agentic settings on InjecAgent and AgentDojo against both open-source and closed-source LLMs (e.g., GPT-4o-mini and GPT-5-nano). Our code is available at https://github.com/albert-y1n/PISmith.

The proposed method (FraudFox) provides solutions to adversarial attacks in a resource constrained environment. We focus on questions like the following: How suspicious is `Smith', trying to buy \$500 shoes, on Monday 3am? How to merge the risk scores, from a handful of risk-assessment modules (`oracles') in an adversarial environment? More importantly, given historical data (orders, prices, and what-happened afterwards), and business goals/restrictions, which transactions, like the `Smith' transaction above, which ones should we `pass', versus send to human investigators? The business restrictions could be: `at most $x$ investigations are feasible', or `at most \$$y$ lost due to fraud'. These are the two research problems we focus on, in this work. One approach to address the first problem (`oracle-weighting'), is by using Extended Kalman Filters with dynamic importance weights, to automatically and continuously update our weights for each 'oracle'. For the second problem, we show how to derive an optimal decision surface, and how to compute the Pareto optimal set, to allow what-if questions. An important consideration is adaptation: Fraudsters will change their behavior, according to our past decisions; thus, we need to adapt accordingly. The resulting system, \method, is scalable, adaptable to changing fraudster behavior, effective, and already in production at Amazon. FraudFox augments a fraud prevention sub-system and has led to significant performance gains.

Cryptocurrency exchanges use proofs of liabilities (PoLs) to prove to their customers their liabilities committed on-chain, thereby enhancing their trust in the service. Unfortunately, a close examination of currently deployed and academic PoLs reveals significant shortcomings in their designs. For instance, existing schemes cannot resist realistic attack scenarios in which the provider colludes with an existing user. In this paper, we propose a new model, dubbed permissioned PoL, that addresses this gap by not requiring cooperation from users to detect a dishonest provider's potential misbehavior. At the core of our proposal lies a novel primitive, which we call Permissioned Vector Commitment (PVC), to ensure that a committed vector only contains values that users have explicitly signed. We provide an efficient PVC and PoL construction that carefully combines homomorphic properties of KZG commitments and BLS-based signatures. Our prototype implementation shows that, despite the stronger security, our proposal also improves server performance (by up to $10\times$) compared to prior PoLs.

Despite the strong multimodal performance, large vision-language models (LVLMs) are vulnerable during fine-tuning to backdoor attacks, where adversaries insert trigger-embedded samples into the training data to implant behaviors that can be maliciously activated at test time. Existing defenses typically rely on retraining backdoored parameters (e.g., adapters or LoRA modules) with clean data, which is computationally expensive and often degrades model performance. In this work, we provide a new mechanistic understanding of backdoor behaviors in LVLMs: the trigger does not influence prediction through low-level visual patterns, but through abnormal cross-modal attention redistribution, where trigger-bearing visual tokens steal attention away from the textual context - a phenomenon we term attention stealing. Motivated by this, we propose CleanSight, a training-free, plug-and-play defense that operates purely at test time. CleanSight (i) detects poisoned inputs based on the relative visual-text attention ratio in selected cross-modal fusion layers, and (ii) purifies the input by selectively pruning the suspicious high-attention visual tokens to neutralize the backdoor activation. Extensive experiments show that CleanSight significantly outperforms existing pixel-based purification defenses across diverse datasets and backdoor attack types, while preserving the model's utility on both clean and poisoned samples.

Authentication is crucial to confirm that an individual or entity trying to perform an action is actually who or what they claim to be. In dynamic environments such as the Internet of Things (IoT), Internet of Vehicles (IoV), healthcare, and smart cities, security risks can change depending on varying contextual factors (e.g., user attempting to authenticate, location, device type). Thus, authentication methods must adapt to mitigate changing security risks while meeting usability and performance requirements. However, existing adaptive authentication systems provide limited guidance on (a) representing contextual factors, requirements, and authentication methods (b) understanding the influence of contextual factors and authentication methods on the fulfilment of requirements, and (c) selecting effective authentication methods that reduce security risks while maximizing the satisfaction of the requirements. This paper proposes a framework for engineering adaptive authentication systems that dynamically select effective authentication methods to address changes in contextual factors and security risks. The framework leverages a contextual goal model to represent requirements and the influence of contextual factors on security risks and requirement priorities. It uses an extended feature model to represent potential authentication methods and their impacts on mitigating security risks and satisfying requirements. At runtime, when contextual factors change, the framework employs a Fuzzy Causal network encoded using the Z3 SMT solver to analyze the goal and feature models, enabling the selection of effective authentication methods. We demonstrate and evaluate our framework through its application to real-world authentication scenarios in the IoV and the healthcare domains.

Robust invisible watermarks are widely used to support copyright protection, content provenance, and accountability by embedding hidden signals designed to survive common post-processing operations. However, diffusion-based image editing introduces a fundamentally different class of transformations: it injects noise and reconstructs images through a powerful generative prior, often altering semantic content while preserving photorealism. In this paper, we provide a unified theoretical and empirical analysis showing that non-adversarial diffusion editing can unintentionally degrade or remove robust watermarks. We model diffusion editing as a stochastic transformation that progressively contracts off-manifold perturbations, causing the low-amplitude signals used by many watermarking schemes to decay. Our analysis derives bounds on watermark signal-to-noise ratio and mutual information along diffusion trajectories, yielding conditions under which reliable recovery becomes information-theoretically impossible. We further evaluate representative watermarking systems under a range of diffusion-based editing scenarios and strengths. The results indicate that even routine semantic edits can significantly reduce watermark recoverability. Finally, we discuss the implications for content provenance and outline principles for designing watermarking approaches that remain robust under generative image editing.

Privacy-Preserving Machine Learning as a Service (PP-MLaaS) enables secure neural network inference by integrating cryptographic primitives such as homomorphic encryption (HE) and multi-party computation (MPC), protecting both client data and server models. Recent mixed-primitive frameworks have significantly improved inference efficiency, yet they process batched inputs sequentially, offering little flexibility for prioritizing urgent requests. Naïve queue jumping introduces considerable computational and communication overhead, increasing non-negligible latency for in-queue inputs. We initiate the study of privacy-preserving queue jumping in batched inference and propose PrivQJ, a novel framework that enables efficient priority handling without degrading overall system performance. PrivQJ exploits shared computation across inputs via in-processing slot recycling, allowing prior inputs to be piggybacked onto ongoing batch computation with almost no additional cryptographic cost. Both theoretical analysis and experimental results demonstrate over an order-of-magnitude reduction in overhead compared to state-of-the-art PP-MLaaS systems.

Apps such as Firechat and Bridgefy have been used during recent protests in Hong Kong and Iran, as they allow communication over ad-hoc wireless networks even when internet access is restricted. However, these apps do not provide sufficient protection as they do not achieve forward secrecy in unreliable networks. Without forward secrecy, caught protesters' devices will disclose all previous messages to the authorities, putting them and others at great risk. In this paper, we introduce FoSAM, the first protocol to provide proven anonymous and forward secret messaging in unreliable ad-hoc networks. Communication in FoSAM requires only the receiver's public key, rather than an interactive handshake. We evaluate the performance of FoSAM using a large-scale simulation with different user movement patterns, showing that it achieves between 92% and 99% successful message delivery. We additionally implement a FoSAM prototype for Android.

Absolute anonymization, conceived as an irreversible transformation preventing re-identification and sensitive value disclosure, has proven to be a broken promise. Modern data protection must therefore shift toward a privacy-utility trade-off grounded in risk mitigation. Differential Privacy (DP) offers a rigorous mathematical framework for balancing quantified disclosure risk with analytical usefulness. Nevertheless, widespread adoption remains limited, largely because complex technical concepts, such as privacy-loss parameters, have yet to be translated into forms meaningful to non-technical stakeholders. This difficulty arises from randomization itself: both analysts and adversaries must draw conclusions from uncertain observations rather than deterministic values. In this work, we adopt an interpretation of the privacy-utility trade-off based on hypothesis testing to measure the uncertainty introduced by randomized mechanisms. In particular, we use the concept of relative disclosure risk to quantify the maximum reduction in uncertainty an adversary can obtain from a membership attack on protected outputs, and show this measure relates directly to standard privacy-loss parameters. We further analyze how DP affects analytical validity via its impact on hypothesis tests assessing statistical significance. Building on these results, we provide practical guidance, accessible to non-experts such as data protection authorities, for navigating the trade-off and selecting protection mechanisms and parameter values.

Watermarking the initial noise of diffusion models has emerged as a promising approach for image provenance, but content-independent noise patterns can be forged via inversion and regeneration attacks. Recent semantic-aware watermarking methods improve robustness by conditioning verification on image semantics. However, their reliance on a single global semantic binding makes them vulnerable to localized but globally coherent semantic edits. To address this limitation and provide a trustworthy semantic-aware watermark, we propose $\underline{\textbf{S}}$emantic $\underline{\textbf{L}}$atent $\underline{\textbf{I}}$njection via $\underline{\textbf{C}}$ompartmentalized $\underline{\textbf{E}}$mbedding ($\textbf{SLICE}$). Our framework decouples image semantics into four semantic factors (subject, environment, action, and detail) and precisely anchors them to distinct regions in the initial Gaussian noise. This fine-grained semantic binding enables advanced watermark verification where semantic tampering is detectable and localizable. We theoretically justify why SLICE enables robust and reliable tamper localization and provides statistical guarantees on false-accept rates. Experimental results demonstrate that SLICE significantly outperforms existing baselines against advanced semantic-guided regeneration attacks, substantially reducing attack success while preserving image quality and semantic fidelity. Overall, SLICE offers a practical, training-free provenance solution that is both fine-grained in diagnosis and robust to realistic adversarial manipulations.

We show that safety alignment in modular LLMs can exhibit a compositional vulnerability: adapters that appear benign and plausibly functional in isolation can, when linearly composed, compromise safety. We study this failure mode through Colluding LoRA (CoLoRA), in which harmful behavior emerges only in the composition state. Unlike attacks that depend on adversarial prompts or explicit input triggers, this composition-triggered broad refusal suppression causes the model to comply with harmful requests under standard prompts once a particular set of adapters is loaded. This behavior exposes a combinatorial blind spot in current unit-centric defenses, for which exhaustive verification over adapter compositions is computationally intractable. Across several open-weight LLMs, we find that individual adapters remain benign in isolation while their composition yields high attack success rates, indicating that securing modular LLM supply-chains requires moving beyond single-module verification toward composition-aware defenses.

Neural Structural Obfuscation (NSO) (USENIX Security'23) is a family of “zero cost” structure-editing transforms (nso\_zero, nso\_clique, nso\_split) that inject dummy neurons. By combining neuron permutation and parameter scaling, NSO makes a radical modification to the network structure and parameters while strictly preserving functional equivalence, thereby disrupting white-box watermark verification. This capability has been a fundamental challenge to the reliability of existing white-box watermarking schemes. We rethink NSO and, for the first time, fully recover from the damage it has caused. We redefine NSO as a graph-consistent threat model within a producer–consumer paradigm. This formulation posits that any obfuscation of a producer node necessitates a compatible layout update in all downstream consumers to maintain structural integrity. Building on these consistency constraints on signal propagation, we present Canon, a recovery framework that probes the attacked model to identify redundancy/dummy channels and then globally canonicalizes the network by rewriting all downstream consumers by construction, synchronizing layouts across fan-out, add, and cat. Extensive experiments demonstrate that, even under strong composed and extended NSO attacks, Canon achieves 100% recovery success, restoring watermark verifiability while preserving task utility. Our code is available at https://anonymous.4open.science/r/anti-NSO-9874.

The rapid evolution of Large Language Models (LLMs) into autonomous, tool-calling agents has fundamentally altered the cybersecurity landscape. Frameworks like OpenClaw grant AI systems operating-system-level permissions and the autonomy to execute complex workflows. This level of access creates unprecedented security challenges. Consequently, traditional content-filtering defenses have become obsolete. This report presents a comprehensive security analysis of the OpenClaw ecosystem. We systematically investigate its current threat landscape, highlighting critical vulnerabilities such as prompt injection-driven Remote Code Execution (RCE), sequential tool attack chains, context amnesia, and supply chain contamination. To systematically contextualize these threats, we propose a novel tri-layered risk taxonomy for autonomous Agents, categorizing vulnerabilities across AI Cognitive, Software Execution, and Information System dimensions. To address these systemic architectural flaws, we introduce the Full-Lifecycle Agent Security Architecture (FASA). This theoretical defense blueprint advocates for zero-trust agentic execution, dynamic intent verification, and cross-layer reasoning-action correlation. Building on this framework, we present Project ClawGuard, our ongoing engineering initiative. This project aims to implement the FASA paradigm and transition autonomous agents from high-risk experimental utilities into trustworthy systems. Our code and dataset are available at https://github.com/NY1024/ClawGuard.

Lightweight block cipher design has largely focused on incremental optimization of established paradigms such as substitution–permutation networks, Feistel structures, and ARX constructions, where security derives from the algebraic complexity of individual components. We propose a different approach based on expander-graph interaction networks, where diffusion and security arise from sparse structural connectivity rather than component sophistication. We present ExpanderGraph-128 (EGC128), a 128-bit block cipher constructed as a 20-round balanced Feistel network. Each round applies a 64-bit nonlinear transformation governed by a 3-regular expander graph whose vertices execute identical 4-input Boolean functions on local neighborhoods. Security analysis combines MILP-based differential bounds, proven optimal through 10 rounds via SCIP, establishing 147.3-bit differential security and conservatively extrapolating to 413 bits for the full cipher. Linear analysis provides MILP bounds of $\geq 2^{145}$, while related-key evaluation shows no free rounds for any nonzero key difference. Additional tests confirm rapid algebraic degree growth and the absence of invariant affine subspaces. Implementation results demonstrate practical efficiency. FPGA synthesis on Xilinx Artix-7 achieves 261~Mbps at 100~MHz using only 380 LUTs, while ARM Cortex-M4F software requires 25.8~KB Flash and 1.66~ms per encryption. These results show that expander-graph-driven diffusion provides a promising design methodology for lightweight cryptography.

AI agents increasingly act through external tools: they query databases, execute shell commands, read and write files, and send network requests. Yet in most current agent stacks, model-generated tool calls are handed to the execution layer with no framework-agnostic control point in between. Post-execution observability can record these actions, but it cannot stop them before side effects occur. We present AEGIS, a pre-execution firewall and audit layer for AI agents. AEGIS interposes on the tool-execution path and applies a three-stage pipeline: (i) deep string extraction from tool arguments, (ii) content-first risk scanning, and (iii) composable policy validation. High-risk calls can be held for human approval, and all decisions are recorded in a tamper-evident audit trail based on Ed25519 signatures and SHA-256 hash chaining. In the current implementation, AEGIS supports 14 agent frameworks across Python, JavaScript, and Go with lightweight integration. On a curated suite of 48 attackinstances, AEGIS blocks all attacks in the suite before execution; on 500 benign tool calls, it yields a 1.2% false positive rate; and across 1,000 consecutive interceptions, it adds 8.3 ms median latency. The live demo will show end-to-end interception of benign, malicious, and human-escalated tool calls, allowing attendees to observe real-time blocking, approval workflows, and audit-trail generation. These results suggest that pre-execution mediation for AI agents can be practical, low-overhead, and directly deployable.

Tool-augmented LLM agents increasingly rely on multi-step, multi-tool workflows to complete real tasks. This design expands the attack surface, because data produced by one tool can be persisted and later reused as input to another tool, enabling exploitable source-to-sink dataflows that only emerge through tool composition. We study this risk as multi-tool vulnerabilities in LLM agents, and show that existing discovery efforts focused on single-tool or single-hop testing miss these long-horizon behaviors and provide limited debugging value. We present ChainFuzzer, a greybox framework for discovering and reproducing multi-tool vulnerabilities with auditable evidence. ChainFuzzer (i) identifies high-impact operations with strict source-to-sink dataflow evidence and extracts plausible upstream candidate tool chains based on cross-tool dependencies, (ii) uses Trace-guided Prompt Solving (TPS) to synthesize stable prompts that reliably drive the agent to execute target chains, and (iii) performs guardrail-aware fuzzing to reproduce vulnerabilities under LLM guardrails via payload mutation and sink-specific oracles. We evaluate ChainFuzzer on 20 popular open-source LLM agent apps (998 tools). ChainFuzzer extracts 2,388 candidate tool chains and synthesizes 2,213 stable prompts, confirming 365 unique, reproducible vulnerabilities across 19/20 apps (302 require multi-tool execution). Component evaluation shows tool-chain extraction achieves 96.49% edge precision and 91.50% strict chain precision; TPS increases chain reachability from 27.05% to 95.45%; guardrail-aware fuzzing boosts payload-level trigger rate from 18.20% to 88.60%. Overall, ChainFuzzer achieves 3.02 vulnerabilities per 1M tokens, providing a practical foundation for testing and hardening real-world multi-tool agent systems.

Textual adversarial attacks pose a serious security threat to Natural Language Processing (NLP) systems by introducing imperceptible perturbations that mislead deep learning models. While adversarial example detection offers a lightweight alternative to robust training, existing methods typically rely on prior knowledge of attacks, white-box access to the victim model, or numerous queries, which severely limits their practical deployment. This paper introduces RTD-Guard, a novel black-box framework for detecting textual adversarial examples. Our key insight is that word-substitution perturbations in adversarial attacks closely resemble the "replaced tokens" that a Replaced Token Detection (RTD) discriminator is pre-trained to identify. Leveraging this, RTD-Guard employs an off-the-shelf RTD discriminator-without fine-tuning-to localize suspicious tokens, masks them, and detects adversarial examples by observing the prediction confidence shift of the victim model before and after intervention. The entire process requires no adversarial data, model tuning, or internal model access, and uses only two black-box queries. Comprehensive experiments on multiple benchmark datasets demonstrate that RTD-Guard effectively detects adversarial texts generated by diverse state-of-the-art attack methods. It surpasses existing detection baselines across multiple metrics, offering a highly efficient, practical, and resource-light defense mechanism-particularly suited for real-world deployment in resource-constrained or privacy-sensitive environments.

Open-domain video platforms offer rich, personalized content that could support health, caregiving, and educational applications, but their engagement-optimized recommendation algorithms can expose vulnerable users to inappropriate or harmful material. These risks are especially acute in child-directed and care settings (e.g., dementia care), where content must satisfy individualized safety constraints before being shown. We introduce SafeScreen, a safety-first video screening framework that retrieves and presents personalized video while enforcing individualized safety constraints. Rather than ranking videos by relevance or popularity, SafeScreen treats safety as a prerequisite and performs sequential approval or rejection of candidate videos through an automated pipeline. SafeScreen integrates three key components: (i) profile-driven extraction of individualized safety criteria, (ii) evidence-grounded assessments via adaptive question generation and multimodal VideoRAG analysis, and (iii) LLM-based decision-making that verifies safety, appropriateness, and relevance before content exposure. This design enables explainable, real-time screening of uncurated video repositories without relying on precomputed safety labels. We evaluate SafeScreen in a dementia-care reminiscence case study using 30 synthetic patient profiles and 90 test queries. Results demonstrate that SafeScreen prioritizes safety over engagement, diverging from YouTube's engagement-optimized rankings in 80-93% of cases, while maintaining high levels of safety coverage, sensibleness, and groundedness, as validated by both LLM-based evaluation and domain experts.

As agentic artificial intelligence systems scale across globally distributed and long lived infrastructures, secure and policy compliant communication becomes a fundamental systems challenge. This challenge grows more serious in the quantum era, where the cryptographic assumptions built into today's AI deployments may not remain valid over their operational lifetime. Here, we introduce quantum secure by construction, or QSC, as a design paradigm that treats quantum secure communication as a core architectural property of agentic AI systems rather than an upgrade added later. We realize QSC through a runtime adaptive security model that combines post quantum cryptography, quantum random number generation, and quantum key distribution to secure interactions among autonomous agents operating across heterogeneous cloud, edge, and inter organizational environments. The approach is cryptographically pluggable and guided by policy, allowing the system to adjust its security posture according to infrastructure availability, regulatory constraints, and performance needs. QSC contributes a governance aware orchestration layer that selects and combines link specific cryptographic protections across the full agent lifecycle, including session bootstrap, inter agent coordination, tool invocation, and memory access. Through system level analysis and empirical evaluation, we examine the trade offs between classical and quantum secure mechanisms and show that QSC can reduce the operational complexity and cost of introducing quantum security into deployed agentic AI systems. These results position QSC as a foundational paradigm for post quantum agentic intelligence and establish a principled pathway for designing globally interoperable, resilient, and future ready intelligent systems.

Suffix jailbreak attacks serve as a systematic method for red-teaming Large Language Models (LLMs) but suffer from prohibitive computational costs, as a large number of candidate suffixes need to be evaluated before identifying a jailbreak suffix. This paper presents Prefix-Shared KV Cache (PSKV), a plug-and-play inference optimization technique tailored for jailbreak suffix generation. Our method is motivated by a key observation that when performing suffix jailbreaking, while a large number of candidate prompts need to be evaluated, they share the same targeted harmful instruction as the prefix. Therefore, instead of performing redundant inference on the duplicated prefix, PSKV maintains a single KV cache for this prefix and shares it with every candidate prompt, enabling the parallel inference of diverse suffixes with minimal memory overhead. This design enables more aggressive batching strategies that would otherwise be limited by memory constraints. Extensive experiments on six widely used suffix attacks across five widely deployed LLMs demonstrate that PSKV reduces inference time by 40% and peak memory usage by 50%, while maintaining the original Attack Success Rate (ASR). The code has been submitted and will be released publicly.

API (Application Programming Interface) keys allow applications to authenticate themselves to third-party services. Inadvertent public exposure of these credentials can pose significant consequences, as adversaries can use them to gain privileged access to other services. In this paper, we measure API credential exposure on the web by analyzing 10M~rendered websites. Our findings reveal that API credential exposure on the web is widespread, affecting organizations such as global banks and core infrastructure providers. We identify 1,748~credentials for accessing 14~providers (e.g., cloud and payment services). Crucially, we demonstrate that these exposures are largely missed by static analysis. By characterizing web-specific exposure vectors and root causes, we find that 62% of JavaScript-based exposures manifest exclusively within compiled deployment bundles, while 16% propagate dynamically through third-party resource inclusions. Moreover, our longitudinal analysis shows these credentials often persist for months to years. We conclude by discussing our responsible disclosure efforts and outlining mitigations to secure web deployment pipelines in the future.

GPUs play an increasingly important role in modern software. However, the heterogeneous host-device execution model and expanding software stacks make GPU programs prone to memory-safety and concurrency bugs that evade static analysis. While fuzz-testing, combined with dynamic error checking tools, offers a plausible solution, it remains underutilized for GPUs. In this work, we identify three main obstacles limiting prior GPU fuzzing efforts: (1) kernel-level fuzzing leading to false positives, (2) lack of device-side coverage-guided feedback, and (3) incompatibility between coverage and sanitization tools. We present cuFuzz, the first CUDA-oriented fuzzer that makes GPU fuzzing practical by addressing these obstacles. cuFuzz uses whole program fuzzing to avoid false positives from independently fuzzing device-side kernels. It leverages NVBit to instrument device-side instructions and merges the resultant coverage with compiler-based host coverage. Finally, cuFuzz decouples sanitization from coverage collection by executing host- and device-side sanitizers in separate processes. cuFuzz uncovers 43 previously unknown bugs (19 in commercial libraries) across 14 CUDA programs, including illegal memory accesses, uninitialized reads, and data races. cuFuzz achieves significantly more discovered edges and unique inputs compared to baseline approaches, especially on closed-source targets. Moreover, we quantify the execution time overheads of the different cuFuzz components and add persistent-mode support to improve the overall fuzzing throughput. Our results demonstrate that cuFuzz is an effective and deployable addition to the GPU testing toolbox. cuFuzz is publicly available at https://github.com/NVlabs/cuFuzz/.