Papers

Model watermarking safeguards AI model intellectual property by embedding distinctive knowledge that induces unique behavioral signatures. The primary technical challenge lies in ensuring watermark robustness against various post-processing attacks on the watermarked model. Model extraction attacks emerge as the most severe threat, where adversaries exploit prediction outputs to train surrogate models that illegally replicate the original model's functionality. In this work, we propose a rehearsal-based watermark embedding framework to enhance the robustness of model watermarks against model extraction attacks. By simulating the extraction process, our method leverages the loss of a simulated stolen model on a trigger set as a training signal to fine-tune the watermark knowledge within the target model. This fine-tuning step encourages the watermark to be embedded in a way that boosts transferability, thereby increasing its chances of persisting and remaining detectable in stolen models. Comprehensive experiments conducted under diverse settings demonstrate that the proposed method significantly improves the robustness of model watermarks against both model extraction and subsequent watermark removal attacks.

This paper explores the value of agentic AI tools for cybersecurity purposes. We evaluate the efficacy of a general-purpose GenAI Large Language Model- (GenAI-) based agent when powered by three different Ollama-hosted general-purpose open source models. We assess each agent's performance using precision, recall, false positive count, and a calculated composite score based upon the interplay of the captured metrics, against the baseline performance of an existing, vetted Static Application Security Testing (SAST) tool, Bandit. Our findings refute the notion that a modern open-source GenAI LLM-based agent is currently suitable for the specialized task of SAST scanning under realistic conditions.

Agent skills let LLM agents reuse instructions, resources, tools, and workflows, but they also create a new place for malicious behavior to hide. A skill may look benign in its documentation or code while becoming harmful only when it is invoked with particular user requests, local assets, persistent state, or multi-step tool interactions. This makes purely static vetting brittle. We present Runtime Skill Audit (RSA), a dynamic analysis method that audits skills by asking what the skill-mediated agent actually does under targeted runtime conditions. Instead of testing every skill with the same generic tasks, RSA profiles risk-relevant interfaces, prepares the execution context needed to exercise them, and assigns security labels from the resulting trace evidence. We instantiate RSA on OpenClaw and evaluate it on 100 skills against representative static baselines. RSA achieves 90.0% accuracy with an 88.0% true positive rate and an 8.0% false positive rate, improving accuracy by 13.0 percentage points over the best static baseline. Under self-evolving attacks, static detectors collapse after one or two rounds, while RSA continues to detect 19–20 out of 20 malicious skills across rounds.

Sybil attacks create an illusion of traffic congestion by utilizing fake identities, which undermines the reliable and safe operation of vehicular ad hoc networks (VANETs). Existing detection mechanisms struggle to effectively handle Sybil attacks as they are (i) susceptible to high false positive rates (FPR) due to the overlapping trajectories of both Sybil and legitimate vehicles, (ii) not practical for real-world deployment due to manual calibrations with ground data, (iii) ineffective for sparse distribution of roadside units (RSUs) and vehicles as they depend heavily on the presence of both, and (iv) inefficient due to computational overheads. This paper addresses these shortcomings and proposes a robust framework to handle Sybil attacks. The proposed scheme reduces the FPR by utilizing Global Positioning System (GPS) location data, enabling the construction of more accurate and distinguishable trajectories. Besides, it employs DBSCAN clustering to identify Sybil vehicles, facilitating unsupervised parameter selection. GPS data eliminates the dependency on RSUs and vehicles, making this scheme effective in both sparse and dense regions. Additionally, the proposed scheme is lightweight and consistent across vehicles with heterogeneous capacities. Experimental results demonstrate that the proposed scheme reduces the FPR by approximately 72% in both dense and sparse regions and lowers the false negative rate (FNR) by 52% in sparse regions. Besides, it exceeds existing works by achieving a detection rate of 99% in sparse regions. Additionally, the proposed scheme decreases the detection time by almost 76% in dense regions and 43% in sparse ones.

Backdoor attacks pose a serious threat to the safety and reliability of Large Language Models (LLMs), as they cause models to behave normally on clean inputs while producing attacker-specified responses when hidden triggers are present. Removing such unknown backdoors is particularly challenging when the defender does not know the backdoor attack types or the internal mechanisms formed through backdoor training. In this work, we propose a simple but effective backdoor removal method based on shared internal mechanisms across different backdoors. First, we show that different backdoors with the same task (attack objective) induce similar trigger-activated changes in the internal activations. Motivated by this observation, our method intentionally embeds a backdoor with a known trigger (dummy backdoor) and then removes it through further fine-tuning on dummy-triggered inputs paired with clean responses. Since the dummy backdoor and the unknown backdoor can rely on shared internal mechanisms, removing the dummy backdoor also reduces the effect of the unknown backdoor. We evaluate our method on three backdoor attack types across multiple model families. Experimental results show that our method substantially reduces the attack success rate of the unknown backdoor while preserving model utility, outperforming representative existing defense methods in both backdoor removal effectiveness and utility preservation. These findings suggest that a defender-controllable backdoor can serve as a helpful proxy for mitigating unknown backdoors in generative LLMs.

Agentic infrastructure introduces a critical control-plane authorization problem: non-deterministic reasoning systems can propose high-stakes mutations to production resources, yet existing security mechanisms – such as identity and access management (IAM), policy engines, consensus protocols, and audit logs – either enforce static, context-unaware permissions or merely record actions post-execution. This paper introduces the Sovereign Assurance Boundary (SAB), a certificate-bound runtime admission layer for autonomous execution authority. SAB intercepts agent proposals at an assurance airlock, compiles them into typed execution contracts $C$, and binds these contracts to cryptographic evidence digests $H(E)$ and policy versions. The contracts are then routed through consequence-aware certification paths. Upon successful admission, the system emits a signed Sovereign Assurance Certificate ($\Omega$) that is strictly scoped to a specific execution identity, revocation epoch, and validity window. Finally, a sovereign execution broker verifies $\Omega$ and performs fresh pre-execution revocation and drift checks before invoking infrastructure APIs. We detail the airlock-broker architecture, formalize its admission and revocation invariants, and report preliminary feasibility measurements from a Go prototype evaluated over 2,500 admission attempts. Ultimately, this broker-enforced model prevents autonomous reasoning from directly mutating state, transforming delegated execution authority into a cryptographically verifiable, evidence-bound, revocable, and replayable runtime artifact.

The widespread adoption of face recognition (FR) technologies raises serious privacy concerns, as facial data can be exploited without consent. To address this challenge, we propose Adv-TGD, a generative adversarial attack framework that synthesizes photorealistic faces capable of impersonating target identities and deceiving face recognition systems. Built upon Stable Diffusion v2.1, Adv-TGD performs per-sample LoRA fine-tuning conditioned on concise textual prompts to generate natural yet adversarially manipulated identities. Unlike conventional identity attack approaches, our method optimizes lightweight cross-attention adapters for each source-target pair within a fixed-timestep denoising process. Latent blending is constrained by a face-local heatmap mask to ensure spatially precise identity manipulation while preserving non-sensitive regions. We introduce a composite objective that integrates masked epsilon-MSE reconstruction, thresholded identity divergence in FR embedding space, directional feature alignment, and source-similarity suppression to balance adversarial attack and visual realism. Optionally, LLaVA-generated attribute prompts enhance fine-grained semantic details without reintroducing identity cues. Under the black-box evaluation protocol, Adv-TGD attains an average attack success rate (ASR) of 85.90% across IR152, IRSE50, MobileFace, and FaceNet, surpassing the semantic SOTA baseline Adv-CPG by 6.25 points, the diffusion-based makeup method DiffAIM by 3 points, and the noise-based P3-Mask by 16 points. Despite its strong attack efficacy, Adv-TGD preserves high visual fidelity (PSNR = 28.18 dB, SSIM = 0.981). Furthermore, we demonstrate the flexibility of our framework by successfully extending it to in-the-wild datasets (LADN), general object classification (ImageNet), and transformer-based diffusion models (FLUX.1).

Collaborative edge-cloud inference enables resource-constrained devices to leverage large language models (LLMs) by offloading partial computation to cloud servers. However, transmitting intermediate activations exposes sensitive user prompts to prompt inversion attacks, where an adversary reconstructs the original input from shared representations. Existing defenses rely largely on heuristic perturbations or empirical tuning, offering limited theoretical understanding of privacy leakage and its interaction with utility and latency constraints. We propose an information-theoretic defense framework for prompt inversion in collaborative LLM inference. Our approach learns privacy-preserving representations by explicitly minimizing the mutual information between intermediate activations and the input prompt while maintaining task utility under computational constraints. We derive theoretical guarantees on prompt reconstruction error, characterize fundamental privacy-utility tradeoffs, and establish token-level accuracy bounds for downstream inference. We then propose a novel defense based on privacy adapters implemented via low-dimensional information bottlenecks. Extensive experiments across multiple settings demonstrate that our method achieves superior privacy-utility-latency tradeoffs compared to existing defenses (up to 35% reduction in attack success), providing a principled foundation for private and efficient collaborative LLM inference.

We study superspace concentration as a quantum resource, formalized through the focus measure F(h̊o}) = {\lambda}_max(h̊o}_super) - the largest eigenvalue of the reduced superspace state - which quantifies the capacity of a quantum system to concentrate informational weight into a preferred subspace of an extended degree-of-freedom space. We develop a complete resource-theoretic framework around this measure and validate its properties through GPU-accelerated numerical simulation. Analytic decoherence predictions are confirmed to machine precision (1.11 x 10^{-16}) for superspace dimensions dS in {2,4,8,16,32}. Focus monotonicity holds across 10,000 random states with zero violations under four focus-non-generating channels across six system configurations. Focused quantum states resist coherent unitary attacks with significantly greater resilience than standard fidelity predicts, with focus remaining above 0.9 at attack strength {\epsilon} = 0.302 versus {\epsilon} = 0.174 for fidelity. We further demonstrate that the focus measure and the U(dS)-asymmetry measure are operationally distinct: asymmetry remains near zero and provides no robustness signal under coherent and targeted attacks while focus tracks spectral concentration and remains robust until {\epsilon} > 0.3. The connection between Grover's algorithm and superspace concentration is made explicit via the identity F(|{\psi}_k><{\psi}_k|) = P(marked), providing a resource-theoretic interpretation of oracle query complexity. Finally, we provide the first numerical characterization of the focus capacity gap {\Delta}F, identifying a log_2(dS) scaling law confirmed for both product and correlated noise channels.

Digital forensic investigations increasingly depend on preprocessing heterogeneous network evidence from intrusion detection systems, IoT devices, and enterprise traffic logs. Incompatible schemas and timestamp formats hinder evidence correlation and timeline reconstruction, while current ad hoc approaches offer no mechanism to verify consistency across runs or analysis, creating reproducibility gaps that challenge evidence admissibility. This paper introduces a deterministic forensic preprocessing framework that converts heterogeneous network datasets into a reproducible canonical form. The framework formalises three preprocessing transformations: schema normalisation, temporal normalisation, and provenance tracking. These transformations are specified using set-theoretic definitions and supported by four theorems establishing determinism, information preservation, and provenance completeness. A chunk-based architecture provides O(c) bounded memory. Empirical evaluation across UNSW-NB15, IoT-23, and TON_IoT demonstrates 100% output consistency across repeated runs, robust temporal normalisation completeness over heterogeneous timestamp formats, and scalable performance from millions to hundreds of millions of records.

Continuous electrocardiography (ECG) monitoring could surface rhythm abnormalities before they escalate into cardiovascular events. However, a deployable system must satisfy three requirements simultaneously: legal-grade privacy (GDPR, HIPAA), real-time inference on constrained edge hardware, and detection quality under non-IID cross-hospital data. We design and evaluate an end-to-end federated system addressing all three for unsupervised 12-lead ECG anomaly detection on PTB-XL dataset, combining three autoencoder families (VanillaAE, ConvAE, VAE), Flower-based federated averaging (FedAvg) across ten simulated hospitals, client-side differentially private SGD (DP-SGD) with a Rényi-DP accountant, and 8-bit integer (INT8) post-training quantization with Raspberry Pi 4 benchmarking. Our main contributions are: an empirical characterization of how these mechanisms compose, practical DP-specific recommendations, and technical and security insights for a clinically sensitive setting. Federated learning matches or exceeds the centralized baseline across all architectures (ConvAE federated area under the ROC curve, AUROC, $0.782$), and an $\varepsilon$ sweep identifies $\varepsilon=4$ as the recommended clinical operating point. INT8 quantization roughly halves model size and cuts Pi 4 latency by up to $44%$ with $<0.12%$ AUROC loss. Crucially, DP and quantization penalties are empirically independent, so practitioners need not trade a strong privacy guarantee for a compact edge footprint. To our knowledge, this is the first system combining federated learning, formal $(\varepsilon,\delta)$-DP, unsupervised reconstruction-based detection, and quantized AArch64 deployment.

Fully homomorphic encryption (FHE) enables computations on encrypted data without decryption, offering strong data privacy at the expense of substantial computational and memory overheads. Prior efforts have steadily improved FHE performance through cryptographic and algorithmic enhancements or hardware acceleration, yet these two directions have progressed largely in isolation, hindering the full exploitation of available hardware capabilities. This work presents WHET, which introduces memory-centric, architecture-aware optimizations to better align cryptographic and algorithmic constructions with FHE accelerator architectures. We identify conventional FHE constructions as major sources of excessive working sets and heavy off-chip memory traffic. We propose accelerator-specific techniques, including fine-grained coefficient-to-slot transformation, plaintext compression, and intermediate modulus raising, to reduce the on-chip data footprint by minimizing temporary ciphertexts and plaintext loads. With these techniques applied, we observe additional opportunities to improve on-chip memory efficiency; hence, we introduce lightweight architectural refinements, including a special-purpose buffer and functional unit extensions. With these optimizations, WHET achieves 1.38-8.74$\times$ per-area performance improvements over state-of-the-art FHE accelerators and the first-ever sub-millisecond CKKS bootstrapping.

Bilateral attribute-based access control for data trading must hide policies, provide cryptographic fairness, and avoid trusted third parties. Existing solutions either leak policy information, incur super-linear costs, or rely on trusted dispute resolution. We present PriME-Deal, a non-interactive protocol that simultaneously achieves policy-hiding bilateral matching, efficient threshold access control, and auditable fair exchange on public blockchains. The seller embeds a secret token under the buyer policy into an oblivious key-value store with pseudorandom masking; the buyer reconstructs the token locally via tag-based probing, eliminating combinatorial enumeration, and proves correctness in zero-knowledge. Fair exchange is enforced through a collateralized on-chain reveal with a cryptographic audit that penalizes misbehaviour without trusted parties. We prove security in the Universal Composability framework under standard assumptions. Compared with the state-of-the-art threshold fuzzy IB-ME scheme, the seller's publishing time is reduced by two orders of magnitude (e.g., 8.76s vs. 690s for a policy of 500 attributes). For a typical configuration of (200,20,5), the buyer completes token reconstruction and proof generation in 8.9s, with the zero-knowledge proof taking under 0.6s and remaining constant across all parameter scales. The on-chain cost is approximately 28.6M gas, well within Ethereum's block limit. PriME-Deal thus delivers the first practical privacy-preserving data trading protocol that combines linear seller overhead, bilateral policy hiding, and auditable fairness.

While private information retrieval (PIR) enables private database services by fully concealing access patterns, it simultaneously requires high computational throughput, large memory capacity, and substantial memory bandwidth. We introduce VIPIR, a versatile GPU framework that co-designs PIR protocols with GPU acceleration. We develop a unified analytic model showing that state-of-the-art PIR protocols fall into two categories with complementary limitations, and propose two protocols that flexibly combine techniques across these categories, overcoming the limitations of both classes. These protocols incorporate a GPU-friendly data compression method called expansion-based ring packing (ExpPack), which offers a high degree of parallelism and minimal communication cost. VIPIR applies further optimizations to core operations, including number-theoretic transforms (NTTs) and various matrix-matrix multiplications (GEMMs). Notably, we develop a tensor-core-based execution method for database multiplication by interpreting it as a mixed-integer-type GEMM. We also design memory-efficient scheduling methods that minimize intermediate buffers and enable multi-GPU scaling under memory capacity constraints. Overall, VIPIR achieves orders-of-magnitude higher throughput than prior PIR systems while reducing communication and memory overheads, making large-scale PIR practical.

As an effective anti-censorship mechanism, network covert channels can provide data privacy protection and ensure communication security. However, the covertness of existing network covert channels primarily depends on the secrecy of their covert algorithms. With the increasing depth of research in this field, the difficulty of breaking such algorithms has gradually decreased. Once the algorithm is exposed, the network covert channel can be easily detected by adversaries. To address this issue, this paper proposes a covert carrier filtering strategy based on the hash. In this strategy, a key-dependent filtering rule is introduced during the construction of the network covert channel, enabling the communicating parties to randomly and dynamically filter a sparse subset from the carrier set as the covert carrier set. This strategy not only enhances the randomness of carrier selection but also tightly couples the covertness of the network covert channel with the security of the key. We employ machine learning-based traffic analysis methods to experimentally validate the strategy in two types of network covert channels: network storage and timing covert channels. The experimental results demonstrate that the proposed strategy significantly improves the detection resistance of network covert channels. When the filter key size exceeds six bits, the impact on the detection effect of the classifier becomes quite significant. Furthermore, the processing delay for a single packet is less than 8 $\mu s$, indicating the feasibility of deploying the proposed strategy in high-speed network environments.

Retrieval-Augmented Generation (RAG) systems enhance large language models by grounding responses in retrieved documents from external knowledge sources at inference time. However, this reliance on retrieved content introduces vulnerabilities to poisoning attacks, in which adversarial documents can manipulate both the retrieval process and the generated outputs. This paper investigates poisoning robustness in RAG through a full factorial experimental study covering 432 configurations. We analyze the impacts of dataset, retriever type, retrieval depth, database composition, chunking strategy, and generator model on retrieval-level and generation-level metrics. The results show that retriever architecture, dataset, and retrieval depth are the strongest factors affecting poisoning exposure, while generator choice and database composition have a major impact on downstream attack success. Dense and graph-based retrievers generally improve robustness relative to BM25, whereas larger retrieval depth increases the likelihood of retrieving poisoned passages. We further show that replicating poisoned content across multiple databases amplifies adversarial influence, while additional clean sources can mitigate it. These findings highlight that poisoning vulnerability in RAG is not attributable to a single component, but instead arises from the interaction of retrieval, generation, and knowledge-base configuration.

Biometric systems are increasingly deployed in security applications; however, they remain vulnerable to spoofing attacks, in which attackers exploit counterfeit biometric data to gain unauthorized access. This research evaluates the effectiveness of state-of-the-art machine learning models, MobileNetV2, DenseNet-121, Inception-v3, and Spoof Trace Disentanglement (STD) in detecting spoofing attacks within facial recognition systems. Using the CelebA-Spoof dataset, the study evaluates model effectiveness using metrics such as accuracy, precision, recall, and F1 Score. Cross-dataset validation is carried out on the MSU-MFSD dataset to assess generalizability. The results show MobileNetV2 as the most efficient model, achieving 92% accuracy while balancing computational effectiveness, making it appropriate for real-life applications. Inception-v3 shows moderate robustness, while DenseNet-121 and STD struggle with generalization. The findings highlight the need for advances in domain adaptation and hybrid architectures to enhance biometric security systems.

The expansion of the digital domain has resulted in a substantial increase in digital communication, with email emerging as one of the most prominent channels. The proliferation of email communication is apparent in both professional and personal contexts, thereby creating numerous vulnerabilities for malicious actors to exploit. Spam emails, a form of unsolicited correspondence often bearing malicious intent towards recipients, have been an ongoing challenge for email users since the inception of email technology, and this problem has been exacerbated by the growth of the digital landscape. Email spam filters are integral components of email clients, engineered to identify potentially harmful messages and alert users to their malicious content. Phishing, frequently the initial phase of malware-based attacks, is evolving rapidly, with malware becoming increasingly sophisticated over time. A widely adopted approach for detecting malicious activity within malware and spam domains is the application of machine learning. Our aim is to assess the impact of the evolution within the spam email domain on these machine learning-based detection systems and to explore strategies for mitigating associated performance degradation.

Jailbreak attacks expose persistent safety weaknesses in large language models (LLMs), but existing stateless single-turn methods face a trade-off: hand-crafted prompts are expressive but static, while iterative prompt optimization can adapt but often relies on low-level mutations that require many target queries. We propose JailbreakOPT, a tool-assisted framework for improving iterative single-turn jailbreak prompt optimization. JailbreakOPT organizes diverse atomic jailbreak prompts into an attack tool library and composes them through a unified intra-episode optimization abstraction to generate stronger standalone attack prompts. To reuse experience across attack episodes, JailbreakOPT further frames tool selection as a contextual bandit problem and applies contextual Thompson sampling to guide exploration and exploitation based on past outcomes. Experiments across multiple target LLMs and attack goals show that JailbreakOPT improves attack success rate (ASR) while reducing the number of attacks until success (No.A) compared with atomic single-turn attacks and existing iterative optimization baselines. This paper may contain offensive or harmful content.

Repository-level benchmarks for evaluating Large Language Model (LLM) code repair on Secure Multi-Party Computation (MPC) software do not yet exist, and directly transplanting general-purpose benchmarks such as SWE-bench fails on three structural fronts: (i) MPC repositories are dominated by generic Python infrastructure rather than cryptographic logic; (ii) high-value MPC fixes lack the standardized tests rigid extraction pipelines require; and (iii) standard fail-to-pass evaluation is insufficient for code that must also be cryptographically safe. MPC is increasingly deployed for privacy-preserving machine learning, biomedical collaboration, and secure analytics. Existing MPC-specific code-synthesis efforts cover only operator-level or single-framework tasks; evaluating LLM agents on real repository-level MPC repair instead demands MPC-aware data curation and a verifier matched to the security and numerical-fidelity guarantees MPC programs must obey neither of which existing benchmarks provide. We introduce MPC-Patch-Bench, a repository-level benchmark organised around two frameworks. (1)The Data Curation Framework combines a domain-specific curation agent that filters raw pull requests through three cryptographic layers with a human-AI completion engine that synthesizes missing problem statements and Fail-to-Pass/Pass-to-Pass tests, yielding 205 fully verified instances. (2)The MPC Verifier provides dedicated security and numerical-fidelity checks via dynamic differential testing against plaintext oracles and MPC-specific static analysis rules that flag unsafe reveals, insecure arithmetic, and illegal public/private casts. The strongest evaluated LLM functionally resolves only 22.9% of MPC-Patch-Bench tasks; the MPC Verifier further reduces verified resolution to 17.1%, with up to 40% of functionally-passing patches rejected for cryptographic or numerical-fidelity violations.

Adversarial robustness evaluations of large language models (LLMs) typically report attack success rate (ASR) under fixed query budgets, implicitly treating all attacks as equally costly. In practice, the computational expense of different attack strategies can vary by orders of magnitude. Consequently, ASR at a fixed budget can obscure the true effort required to jailbreak a model, thereby making it hard to determine whether an attack's cost justifies its payoff to the attacker. We propose a compute-aware evaluation framework based on computational pressure, measured in cumulative floating-point operations (FLOPs), as a proxy for adversarial effort. We introduce risk-compute curves, which map compute budgets to attack risk, and derive two metrics that summarize the average pressure required for a given attack to succeed. Across ten models spanning three families and four different stages in language model training and alignment, evaluated with three attack strategies (gradient-based, iterative refinement, and template-based) on two jailbreak robustness benchmarks, we find: (1) alignment training has non-monotonic effects on compute-space robustness; (2) scaling model size reduces gradient-based attack effectiveness but has limited impact on cheaper template-based attacks; (3) gradient-based attacks optimized on a surrogate model can transfer to a separate target model, providing a way to reduce attacker costs; (4) compute cost varies by up to ${\approx}5{\times}$ across harm categories within a single model; and (5) safety-aligned RL increases aggregate cost while leaving some categories disproportionately accessible. We release our framework to enable compute-aware risk assessment and evaluation.

Data leakage – contamination of a model with information unavailable at baseline – is the dominant reproducibility failure in machine-learning-based science, yet detection tools require training code, external data, or domain expertise. None operates on the artifact an auditor most often holds: the model's output. We ask what can be decided about leakage from predictions and outcomes alone. We give a decision-theoretic framework in which leakage diagnostics are functionals of the predicted-risk/outcome law, parameterized by a threshold-weighting linked to proper scoring rules and decision-curve analysis. We prove a sharp impossibility: a recalibrated leak matching an honest model's calibration and discrimination is indistinguishable from honest performance by any function of the predictions, so the broad class is detectable only against an externally supplied ceiling on achievable discrimination. We then prove what leakage cannot hide: a near-deterministic subgroup – the signature of a near-label leak – produces a sustained unit-purity head that no legitimate predictor of a non-deterministic outcome can manufacture, yielding a prior-free test. These results organize leakage into a trichotomy – miscalibrated, broad-calibrated, and deterministic – each with a matched detector and failure mode. We validate on UK Biobank using time-windowed comorbidity leakage with known, graded severity, measuring a detection floor of $\Delta\cstar \approx 0.007$ on this endpoint, below which residual leakage is undetectable from output and too small to alter conclusions. The numerical floor is cohort- and endpoint-specific; the structural lesson is general: output-only detection fails where residual leakage is indistinguishable from an honestly stronger predictor. The test returns a verdict on a prediction vector in under a second on commodity hardware.

Retrieval-Augmented Generation (RAG) systems are vulnerable to corpus poisoning attacks that manipulate downstream model outputs through malicious knowledge injection. Existing studies mainly evaluate poisoning under simplified retrieval settings, overlooking practical RAG pipelines involving document chunking, dense retrieval, reranking, and grounded generation. In this paper, we revisit corpus poisoning under realistic multi-stage retrieval pipelines and show that many existing attacks substantially degrade after reranking despite achieving high retrieval-stage relevance. We identify retrieval granularity mismatch as a key reason for this failure: document-level adversarial signals are often fragmented during chunking, while rerankers favor locally coherent and answer-bearing passages rather than globally optimized semantic similarity. Based on this observation, we propose Chunk-aware and Rerank-Consistent Poisoning (CRCP), a poisoning framework that jointly optimizes retrieval relevance, reranker consistency, and chunk-boundary robustness. CRCP explicitly models chunking transformations during optimization to generate locally self-contained adversarial passages that remain effective under varying chunking configurations. Experiments on standard RAG benchmarks with multiple retrievers and rerankers show that existing poisoning methods are highly sensitive to chunk size and reranking strategies, whereas CRCP achieves substantially higher attack success rates and stronger robustness across realistic retrieval pipelines. Our findings highlight an important realism gap in current RAG security evaluation and suggest that poisoning in modern RAG systems should be studied as a multi-stage retrieval consistency problem rather than a retrieval-only problem.

Small Office/Home Office (SOHO) devices are widely popular, yet often attacked due to security vulnerabilities in their firmware, affecting thousands of devices. These security vulnerabilities often stem from outdated Linux kernel versions included in SOHO device firmware. Naturally, prior work audited the extent and impact of this issue by simple Linux version extraction and version number based vulnerability mapping. However, it is unclear how many of these anticipated vulnerabilities actually exist in the heavily customized SOHO kernels and if there are any barriers towards updating Linux kernels in SOHO firmwares. To address this gap, we uncover actual kernel-related vulnerabilities found in 306 SOHO devices using a high-precision template-based CVE detection mechanism on GPL source releases of more than 900 firmwares from these devices. Next, as a first, we traced the supply chain of these vulnerable SOHO devices at scale and identify kernel lock-in as a significant security issue – SOHO vendors are effectively locked to specific (often older) kernel versions due to the system-on-chip (SoC) SDKs they use. This kernel lock-in produces a vulnerability debt that is inherited along the supply chain from SoC vendor to firmware creators (ODM/OEM) to router/IP-camera vendor and ultimately borne by end users. All five SoC vendors in our dataset had used SDKs with Linux kernels that had reached EoL more than a year before their usage in a SOHO device. Finally, we explore the mitigation-potential of individual, regulatory and community governance by analyzing social media posts, regulations and community efforts. Our results show that regulation compliance is insufficient and only SoC vendors who engage with communities for kernel upgradation offered a viable path towards mitigation. The data and code for this work is available at https://doi.org/10.5281/zenodo.20433799

Generative AI applications such as personal AI agents, image generators, and chat assistants offer advanced capabilities to improve user experience. Behind the scenes, Large Language Models (LLMs) that power these services require a massive amount of computation and are usually deployed in the cloud, available as APIs, meaning that a user's request has to be sent to a Cloud Inference Service (CIS) for processing. However, the strong capabilities of LLM also mean that user's requests now contain much more personal sensitive or enterprise confidential information, demanding equally strong protection in CIS. While early industry efforts such as Apple Private Cloud Compute (PCC) and Google Private AI Compute have emerged to show the potential of secure CIS, they are not adoptable for deployment by others due to their reliance on proprietary hardware and closed ecosystem. In addition, they all suffer from their own design glitches that can undermine the ambitious goal of bringing in true privacy protection to end users. In this paper, we present our analysis of the fundamental requirements of building a secure yet open CIS. We then present OpenPCC, a Confidential CIS framework that does not rely on proprietary hardware but instead uses commercially available TEEs. We implement an open-source prototype and characterize it end-to-end on a Llama-3 8B vLLM workload, separating OpenPCC's own cost from the underlying TEE hardware. Our analysis and evaluation demonstrated the feasibility and security of the system.

We present a longitudinal study of approximately 1.52 million malicious domains observed on VirusTotal (VT) between January and May 2026. Domains were selected on the basis of detection by at least five independent VT scanning engines and a first-seen date within the study window. We group the dataset into compromised domains and attacker created domains, which account for approximately 89.3% of the dataset. Combining WHOIS registration records and passive DNS (PDNS) data with the VT dataset, we characterise attacker behaviour across eight dimensions: temporal distribution, compromisedvs.attack classification, domain age at first detection, registrar and TLD preferences, DNS query volume as a damage proxy, hosting infrastructure concentration (IP and ASN level), bulk registration patterns, and brand impersonation. Key findings include: the majority of attacker created domains are short lived registrations used within weeks of creation; a small number of registrars and TLDs account for most abuse; Cloudflare infrastructure is heavily exploited for domain fronting; bulk registration events involving thousands of domains from a single registrar on a single day are widespread; and several global brands, particularly WhatsApp and Google, are heavily impersonated. We share the annotated dataset in the GitHub repo https://github.com/mufimash/malicious_domains for further research.

Recent deep learning approaches for network intrusion detection increasingly incorporate temporal architectures such as recurrent networks and Transformers, often reporting near-perfect performance on CIC-IDS2017. However, many existing studies neither supply their temporal modules with genuine sequence inputs nor evaluate under realistic, leakage-free conditions, making it unclear whether reported gains arise from true sequence-modelling capability. In this work, we reformulate CIC-IDS2017 as a temporal intrusion-detection task by constructing ordered flow sequences from network conversations and benchmarking nine classical and deep learning architectures under a random split, two leakage-free splits, and a padding-scheme ablation. The central finding is that padding convention, not architecture, determines the Transformer's performance: on genuinely sequential (non-padded) windows the Transformer achieves the highest macro-F1 of any model in the experiment ($0.89$), yet under zero-pad+mask evaluation it drops markedly while LSTM, GRU, and 1D-CNN remain stable and a class-balanced Random Forest is the most robust model overall. The same repeat-last padding cue masks a $67\times$ increase in the Transformer's false-alarm rate that only leakage-free evaluation reveals. Evaluation methodology – specifically padding convention and split protocol – thus has a larger effect on reported performance than architectural choice. We advocate for leakage-free splits, explicit padding disclosure, and sequence-aware benchmarking as standard practice in future IDS research. Our code and implementation details are available here: https://github.com/zachmocz/temporal-ids-bench.

Advanced AI systems for code analysis, binary analysis, fuzzing orchestration, and penetration-test planningmay significantly increase the rate at which latent vulnerabilities are discovered. While improved discovery can benefit defenders, it can also overload remediation pipelines and accelerate adversarial weaponization. This paper develops a queueing and network-theoretic model of AI-accelerated vulnerability discovery in interconnected systems. We represent an enterprise as a weighted dependency graph with replenishing vulnerability pools, finite remediation capacity, triage degradation, exploit window compression, and dynamic compromise propagation. We derive stability conditions for vulnerability backlogs, formulate a dynamic coupling between unresolved backlog and cascade risk, and evaluate mitigation strategies through simulation. Results indicate that when actionable discovery arrivals exceed remediation throughput, backlogs grow rapidly and systemic risk increases nonlinearly. In hub-dominated topologies, segmentation can reduce propagated compromise more effectively than remediation speed alone, while the strongest defense combines remediation automation with reduced network coupling.

OpenClaw has rapidly emerged as a transformative artificial intelligence (AI) agent framework, and its ability to autonomously execute complex, multi-step tasks has attracted an ever-growing and diverse user base. However, this capability comes with significant risks. While existing research has made important strides in characterizing these threats, such work is predominantly directed at technically sophisticated audiences. It remains largely inaccessible to non-technical users. This demographic now makes up an increasingly large and underserved portion of the community, yet it is these very users who most urgently need practical and straightforward guidance. In response, we bridge this gap through a series of interconnected efforts designed to lower the risk barrier for non-technical OpenClaw users. First, we identify and categorize seven core risks that OpenClaw users may encounter in daily usage, explaining each in plain language so that non-technical users can readily grasp the nature and potential consequences of these threats. Second, for each identified risk, we distill a set of corresponding defensive strategies into clear and actionable operational steps that are easy to follow. Third, to make protection even easier, we provide a companion OpenClaw Skill that automates key security configurations, enabling users to safeguard their systems with minimal manual intervention. Through this work, we demonstrate that safeguarding against the risks of intelligent agents need not be the exclusive domain of security experts, and that non-technical users can meaningfully participate in reducing these risks through simple, practical actions.

AI-powered code generation systems have transformed software development but introduce critical inference-time security vulnerabilities. This research presents a systematic investigation of context-based adversarial attacks, where strategically crafted contextual inputs, including comments, documentation, variable names, bias large language models toward generating exploitable code. Through 2,800 controlled experiments across CodeT5+, CodeLlama, GPT-3.5-Turbo, and GPT-4, we quantify attack effectiveness and defense mechanisms. Results demonstrate that adversarial conditions increase vulnerability generation 10.7x (from 3.5% to 37.4%), with direct instruction attacks achieving 100% success on GPT-3.5-Turbo. Cross-model transferability reaches 60-100%, indicating systemic architectural vulnerabilities rather than model-specific flaws. Our dual-layer defense framework achieves 89.1% detection rate with 0.3% false positives and 520ms latency, demonstrating practical feasibility for real-time deployment in development environments.