Papers

Backdoor attacks threaten the deep-learning supply chain by poisoning a small fraction of the training data so that a model behaves normally on clean inputs but maps triggered inputs to an attacker-chosen class. Clean-label backdoor attacks are especially difficult to audit because poisoned examples preserve their semantic labels. Yet existing clean-label attacks often require surrogate model training, auxiliary data access, or iterative optimization. In this paper, we present Checkerboard, a clean-label backdoor attack with closed-form, data-independent trigger design. We formulate trigger design through an input-space Fisher-separability objective and, under a ridge four-neighbor local-smoothness prior for natural images, obtain the pixel-wise checkerboard as a closed-form maximizer of the resulting design proxy without data access, model training, or optimization. Across four benchmark datasets, Checkerboard outperforms the evaluated norm-bounded clean-label attacks and achieves state-of-the-art performance under low global poisoning rates. For example, on CIFAR-10, under a trigger perturbation of $10/255$, poisoning 20 training samples achieves $95.72\%$ Attack Success Rate (ASR). On IN-100, a global poisoning rate of only $0.4\%$ yields over $83\%$ ASR without degrading clean accuracy. The proposed attack also remains effective against state-of-the-art backdoor defenses and shows resistance to adaptive defenses under simple modification.

AI browsing agents are an emerging class of AI-powered bots capable of autonomously navigating websites. Unlike traditional web bots, AI browsing agents typically operate using real browsers and perform everyday tasks, making them difficult to detect. Yet little is known about whether existing AI browsing agents can be distinguished from humans and one another based on their browser or behavioral fingerprints. In this paper, we present the first controlled measurement study of seven AI browsing agents and human users. Using an instrumented honey website, we collect browser and behavioral fingerprint features while AI browsing agents and humans perform three tasks: flight booking, online shopping, and forum interaction. We then train FP-Agent, a multi-class classifier, to evaluate the discriminative power of these features. We find that browser fingerprints provide limited discriminative power when shared by multiple AI browsing agents. Behavioral fingerprints, however, are distinctive: differences in typing, scrolling, and mouse behavior separate AI browsing agents from humans and one another. In a case study evaluating Cloudflare's bot detection, FP-Agent detects all seven AI browsing agents, whereas Cloudflare detects only one. Our findings show that behavioral fingerprints are a critical component to reliably detect and control this emerging form of web traffic.

Account-based ledgers – standard externally-owned accounts (EOAs), ERC-4337 smart accounts, post-Pectra EIP-7702 delegated EOAs – place the holder of the controlling key at the apex of asset authorization. We ask a structural question about ledger access control: under this authorization model, can a protocol enforce the future disposition of an asset without taking custody and without requiring the owner's cooperation at enforcement time? We formalize the target as Non-Custodial Enforced Encumbrance (NCEE), a four-property specification covering self-custody, transition restriction, irrevocability, and permissionless enforcement. We define the Key Sovereignty Axiom (KS) and prove that any ledger satisfying KS cannot realize NCEE; standard EOAs, ERC-4337 smart accounts, and EIP-7702 delegated EOAs satisfy KS for their standard asset paths. We define Asset-Authorization Coupling (AAC) and prove it necessary for NCEE in the transfer-dichotomous asset setting. To witness the positive side, we introduce the envelope, a primitive for commitment-based private-state ledgers that binds a note, a condition tree, and a redistribution intent to protocol-maintained marker sets, separating ordinary spend nullifiers from a new encumbrance-namespace nullifier derived from note randomness rather than the owner key. We prove the envelope realizes NCEE under stated cryptographic assumptions and a deployment assumption that the marker-set registry is immutable; three concrete deployment templates are given. We define games for encumbrance integrity, settlement security, key-compromise resilience, and encumbrance indistinguishability. A reference implementation in Noir and UltraHonk supports the empirical claims, with gas measurements, recursive aggregation benchmarks, and a practical-economics analysis.

Blockchain and decentralized finance have revolutionized the financial ecosystem while simultaneously exposing it to cryptocurrency phishing attacks. Existing phishing detection methods primarily rely on graph learning, but they face significant limitations. Static graph learning approaches fail to account for the temporal evolution of phishing patterns, while semi-dynamic methods, such as those combining static GNNs with LSTM, struggle to capture the irregular and bursty nature of blockchain transactions. Moreover, these methods overlook the diversity of Ethereum transactions, treating them as homogeneous graphs, and heavily rely on supervised learning, which requires extensive labeled data that is not readily available. These limitations reduce their adaptability to emerging phishing threats. In this paper, we present PhishEye, a fully dynamic self-supervised system that monitors on-chain transactions to detect phishing activities. PhishEye formulates Ethereum transactions as a heterogeneous temporal attributed multi-graph and incorporates a novel temporal graph contrastive learning model, which captures both temporal patterns and heterogeneous transaction types. The evaluation on a dataset of 161,658 addresses and 416,541 transactions shows that PhishEye outperforms existing methods, achieving an F1 score of 87.23% and an AUC of 98.43% for phishing transaction detection, and an F1 score of 94.19% and an AUC of 98.03% for phishing account detection. In real-world deployment from May 1, 2023 to July 31, 2024, PhishEye identified 1,803 previously unknown phishing addresses, providing early alerts that helped prevent losses exceeding 2 billion USD.

In this paper, we propose a method for privacy-preserving federated learning that uses randomly selected model parameters to update global models. High-quality deep neural networks (DNN) models require a huge amount of training data in general, but model training raises privacy concerns when dealing with sensitive or personal information. Federated learning is a distributed machine learning framework in which multiple clients and a server train a model collaboratively. However, if the shared updates are compromised, an attacker may reconstruct the original training data. In addition, previous methods for improving robustness generally reduce the accuracy. To overcome these issues, in our method called federated learning using randomly selected model parameters (FLRSP), model parameters computed in each local server are randomly selected and shared to update a global model in a central server. In experiments, image classification tasks were carried out on the ResNet34 architecture and the Vision Transformer (ViT) under the use of Federated Stochastic Gradient Descent (FedSGD) and Federated Averaging (FedAvg), and the results demonstrated our method's effectiveness in terms of image classification accuracy and robustness against state-of-the-art attacks compared with previous methods.

AI-driven penetration testing agents are now capable of autonomously executing attacks within compromised networks. Identifying the model family that controls the active sessions of such agents provides valuable information towards understanding the intent of the attack and further developing attack countermeasures. In this paper, we introduce Trace, a novel multi-stage attribution and forensic framework for AI attack agents using terminal command sequences. Once Trace identifies a model family for the attacker agents, it guides a defensive prompt injection (DPI) strategy to the attacker model via a crafted payload. This is with the aim to exfiltrate system prompts from an attacker model, thus, revealing valuable information to understand the attacker intent and facilitate further forensic investigation. We have implemented our approach revolving around a Linux capture-the-flag (CTF) box. The attacker agents are bolstered via three distinct scaffolds and seven frontier model families. Our evaluation reveals that Trace achieves a macro F1 score of 0.981 in accurately fingerprinting the attacker model family (0.815 when generalizing to unseen scaffolds). Besides, the fingerprinting guides the DPI via a crafted payload to certain model families, resulting in system prompt extraction from 81.9% of non-Claude sessions on average (up to 98.3%) at 0.736 Sentence-BERT fidelity – 1.88x higher than blind deployment. Finally, to validate the robustness of Trace, we evaluate it with a blackbox and proprietary scaffold employing multiple model families (Gemini and Claude Opus). Our evaluation identified the model family with an average 78% accuracy. Moreover, for the Gemini model family, the DPI employed by Trace revealed the entire system prompt and this has been confirmed by the developers. Trace therefore provides a fundamental first step towards attacker agent forensics.

Low Earth Orbit (LEO) satellite constellations are evolving from communication relays into autonomous platforms operating in increasingly congested and contested environments. Since uplinks to ground stations can be severed or jammed, ensuring reliable coordination among satellites requires autonomous Byzantine Fault-Tolerant (BFT) consensus. However, applying conventional BFT protocols to LEO constellations is challenging due to their dynamic topology, sparse connectivity, and limited communication bandwidth. In this paper, we present OrbitBFT, a novel two-stage hierarchical BFT consensus protocol tailored to the unique characteristics of LEO constellations. First, OrbitBFT exploits the topological stability within orbital planes to partition the constellation and perform localized intra-plane consensus, which reduces communication overhead. Second, we design a Byzantine-resilient bypass mechanism and a hop-by-hop transmission protocol to ensure reliable message delivery and mitigate congestion, even in the presence of adversarial behavior. Third, we adapt and optimize PBFT and HotStuff to the LEO context, achieving linear message complexity while preserving safety and liveness. Extensive evaluations in a realistic Starlink-based simulation demonstrate that OrbitBFT significantly improves scalability, throughput, and latency compared to its original designs, making it a practical and efficient BFT solution for large-scale satellite networks.

Metric differential privacy (mDP) strengthens local differential privacy (LDP) by scaling noise to semantic distance, but many machine learning (ML) systems are consumed under joint observation, where model-agnostic, per-record guarantees can miss leakage from evidence aggregation. We introduce metric-normalized posterior leakage (mPL), an attacker-aligned, distance-calibrated measure of posterior-odds shift induced by releases, and show that for single or independent releases, uniformly bounding mPL is equivalent to mDP. Under joint observation, however, satisfying mDP may still leave mPL high because learned aggregators compound evidence across correlated items. To make control practical, we formalize probabilistically bounded mPL (PBmPL), which limits how often mPL may exceed a target budget, and we operationalize it via Adaptive mPL (AmPL), a trust-and-verify framework that perturbs, audits with a learned attacker, and adapts parameters (with optional Bayesian remapping) to balance privacy and utility. In a word-embedding case study, neural adversaries violate mPL under joint consumption despite per-record mDP perturbations, whereas AmPL substantially lowers the frequency of such violations with low utility loss, indicating PBmPL as a practical, certifiable protection for joint-consumption settings.

Large language model (LLM)-powered multi-agent systems (MAS) enable agents to communicate and share information, achieving strong performance on complex tasks. However, this communication also creates an attack surface where malicious agents can propagate misinformation and manipulate group decisions, undermining MAS safety. Existing embedding-based defenses aim to detect and prune suspicious agents, but their effectiveness depends on a clear separation between the text embeddings of malicious and benign messages. Attackers can circumvent such defenses by crafting messages whose embeddings lie close to benign ones. We analyze this failure mode theoretically and validate it empirically with three attacks, Slow Drift, Benign Wrapper, and Chaos Seeding. Our analysis further reveals a fundamental limitation of embedding-based defenses: because they rely solely on the text embeddings, they ignore token-level confidence signals such as logits, which can remain informative when embeddings are not distinguishable under attack. We propose using confidence scores to prune or down-weight messages during MAS communication. Experiments show improved robustness across models, datasets, and communication topologies. Moreover, we find that the effectiveness of confidence signals decays over communication rounds, highlighting the importance of early intervention. This insights can inform and inspire future work on MAS attacks and defenses.

Machine unlearning (MU) has emerged as a key mechanism for ensuring data privacy and regulatory compliance by enabling models to forget specific training samples. However, recent studies have shown that the removal of data can inadvertently introduce privacy leakages to the retain set,i.e., data that remain in the model after unlearning. In this paper, we extend the scope of privacy analysis in unlearning to the often-overlooked retained data. We introduce TC-UMIA, the first tri-class unlearning membership inference attack. TC-UMIA is a population-level inference framework that leverages model predictions before and after unlearning to distinguish among the forget, retain, and unseen set. Extensive experiments on five state-of-the-art unlearning algorithms and six real-world datasets demonstrate that: (i) unlearning can introduce additional privacy risks to the retain set, making it more susceptible to membership inference attacks; (ii) TC-UMIA is effective across a wide range of model architectures, datasets, and MU approaches. Beyond launching the attack, we rigorously evaluate three defense mechanisms, namely label-only outputs, dropout, and differential privacy, to mitigate the privacy risks posed by TC- UMIA. Our results reveal a fundamental trade-off between privacy protection and model accuracy, with the dropout approach offering the most favorable balance.

Retrieval-augmented generation and tool-integrated LLM agents increasingly depend on external textual sources. This reliance broadens the available attack surface, allowing adversaries to insert malicious instructions that trigger unintended model behaviors. Current defensive measures often utilize LLM-based detectors to filter such content, but these approaches remain vulnerable to optimization-based attacks. Additionally, training-based methods frequently fail to generalize to novel data distributions. To resolve these issues, we introduce SONAR, a prompt sanitization framework that identifies and removes injected content using metrics from natural language inference. Specifically, SONAR constructs a sentence-level relational graph across the user query and external data. By using entailment and contradiction scores as edge weights, the system identifies sentences that deviate from the core task. It then employs connectivity-driven pruning to eliminate flagged injection seeds and their related neighbors while maintaining benign context. Rigorous evaluations across several models and datasets show that SONAR reduces the attack success rate to nearly zero, significantly outperforming nine established baseline defenses.

Hallucinations remain an unsolved problem for LLMs, and package hallucinations are a particularly dangerous instance of this phenomenon. Package hallucinations occur during code generation when a model fabricates non-existent software packages, recommending imports and installation commands for fictional libraries. This creates a critical supply-chain vulnerability; an attacker can proactively register such packages on public registries with malicious payloads that are subsequently installed and executed by developers or autonomous agents. These hallucinations enable a class of package confusion attack known as slopsquatting. To address this issue, we present Adaptive Unlearning (AU), a post-deployment framework that surgically suppresses package hallucinations while preserving general model utility. AU introduces a hybrid token-level objective that simultaneously reinforces valid outputs and suppresses hallucinated ones. Combined with an adaptive discovery loop that continuously surfaces new hallucination-inducing contexts without human supervision, AU enables generalization to unseen prompts and hallucinations. We demonstrate that AU reduces package hallucination rates by 88%, while maintaining performance on standard coding benchmarks. Our analysis shows that distributional changes are concentrated on package-related generations, leaving general coding behavior largely unaffected and confirming that AU's effect is isolated to the targeted distribution. AU relies entirely on model-generated data and requires no human annotation, representing a post-deployment hallucination mitigation framework.

We present a certified purity architecture that converts governance enforcement in cognitive workflow systems from a runtime convention into a structural capability boundary. A prior three-layer governance architecture proves governance completeness, provenance completeness, and the impossibility of ungoverned effects, conditional on the pure module constraint: that step executors cannot perform effects. That constraint was enforced by module import graph analysis, which is insufficient against adversarial bypass on the BEAM virtual machine. This paper closes the gap through four mechanisms: (1) a restricted WebAssembly compilation target where effect-producing instructions are structurally absent; (2) purity certificates, cryptographically signed proofs binding executor binaries to their import classifications; (3) a runtime verification gate that rejects uncertified executors before they enter the governance pipeline; and (4) portable governance credentials via remote attestation for cross-organizational verification. We prove four theorems: structural purity by construction, bypass elimination for all five BEAM bypass classes, certificate integrity, and gate completeness. The guarantee holds relative to an explicit Trusted Computing Base. Evaluation on four implemented executors shows verification latency of 39–42 us, full plan cycle under 400 us, runtime overhead under 0.4% of a 100 ms HTTP request, and zero determinism divergences across repeated invocations.

Shor-style quantum algorithms for the elliptic-curve discrete logarithm problem (ECDLP) are highly sensitive to the exact semantics of their group-operation oracles. Consequently, minor implementation choices can invalidate the intended mathematical model and lead to misleading conclusions. This paper introduces a semantics-first verification perspective for an end-to-end, compilable ECDLP implementation built on Qrisp. We specify the implemented oracle at the level of program semantics, derive refinement-style verification obligations for its key components, and provide a high-level complexity argument for the resulting oracle family. A small case study highlights that (i) the core point-update primitive agrees with a classical reference on well-formed inputs, yet (ii) controlled execution may violate the expected control law under the evaluated toolchain, despite a passing trivial control sanity check. These results position semantic auditing as a practical prerequisite for trustworthy ECDLP-oriented quantum software.

LLMs are increasingly equipped with safety alignment mechanisms, yet recent studies demonstrate that they remain vulnerable to jailbreaking attacks that elicit harmful behaviors without explicit policy violations. While a growing body of work has explored automated jailbreak strategies, existing methods face several fundamental challenges, including the lack of systematic utilization of both successful and failed attack experiences, as well as the absence of principled mechanisms for composing and selecting reusable attack rules under diverse constraints. As a result, existing methods struggle to accumulate transferable knowledge over time and to reliably adapt attack strategies across different targets and evolving safety mechanisms. To address these issues, we propose a Self-Evolving Rule-Driven Training-Free Jailbreak (SRTJ) framework that systematically discovers, composes, and refines attack strategies through interaction and feedback, without updating model parameters. Specifically, SRTJ couples experience-driven attack generation with answer set programming (ASP)-based rule selection and constraint-aware composition, where iterative verifier feedback is leveraged to jointly refine successful strategies and analyze failure patterns. The resulting rule memory evolves in a hierarchical multi-level manner, explicitly organizing distilled attack knowledge into long-term, middle-term, and short-term rules, thereby capturing both stable transferable strategies and transient adaptive behaviors to effectively balance exploration and exploitation across attack attempts. Extensive experiments on mainstream jailbreak benchmark (HarmBench) demonstrate that SRTJ achieves strong and stable attack performance across different target LLMs, while exhibiting improved robustness and generalization compared to existing jailbreak methods. The code is available at https://github.com/TheSolkatt/SRTJ.

We develop a unified mathematical formulation for post-quantum authenticated telemetry and actuation in FADEC-coupled dual-spool turbofan cyber-physical systems. The formulation integrates lattice-based key establishment under LWE/SIS-style assumptions, PUF-derived attestation entropy, authenticated encryption, radar-altimeter integrity, avionics-bus timing, and Kalman residual monitoring in a stochastic hybrid model. Within this model, plant evolution, communication latency, leakage, adversarial channel quality, and cryptographic state evolve under a common filtration. We show that channel uncertainty tightens admissible key-renewal periods, that ciphertext expansion enters bus-level schedulability constraints, and that sensing and actuator limits shape integrity thresholds and allowable control delay. We further relate PUF smooth min-entropy to distinguishing advantage and connect innovation statistics to conservative alarm design. Overall, the results characterize how post-quantum security, real-time schedulability, and closed-loop stability interact in safety-critical aerospace control architectures within a defensive analytical treatment that does not provide operational guidance for interference with real platforms.

Retrieval-Augmented Generation (RAG) equips large language models (LLMs) with external evidence by retrieving documents at inference time, but it also turns the retrieval corpusinto a sensitive asset. Under a black-box setting, an adversary given a candidate document can infer whether it has been ingested into the RAG knowledge base (i.e., document-level membership inference) solely from query response interactions, thereby leaking corpus coverage and the existence of sensitive topics. Existing RAG MIA methods either rely on soft signals such as semantic similarity, which often yield overlapping member/non-member score distributions and unstable thresholds, or employ explicit confirmation probes whose intent is conspicuous and thus prone to refusal and detection. We propose E-MIA, which converts verifiable hard evidence in the target document (e.g., fine-grained details, proper nouns/technical terms, definitional statements, metadata cues, and causal/constraint relations) into an exam with four objectively gradable question types (FB/SC/MC/T/F), and uses the aggregated exam score across multiple evidence targeted questions as the membership signal. Experiments across multiple datasets and diverse RAG configurations demonstrate that E-MIA improves member/non-member separability in stringent settings while preserving natural, stealthy queries, and we further analyze the impact of question composition and exam length on attack effectiveness.

Background: Patient-facing medical chatbots based on retrieval-augmented generation (RAG) are increasingly promoted to deliver accessible, grounded health information. AI-assisted development lowers the barrier to building them, but they still demand rigorous security, privacy, and governance controls. Objective: To report an anonymized, non-destructive security assessment of a publicly accessible patient-facing medical RAG chatbot and identify governance lessons for safe deployment of generative AI in health. Methods: We used a two-stage strategy. First, Claude Opus 4.6 supported exploratory prompt-based testing and structured vulnerability hypotheses. Second, candidate findings were manually verified using Chrome Developer Tools, inspecting browser-visible network traffic, payloads, API schemas, configuration objects, and stored interaction data. Results: The LLM-assisted phase identified a critical vulnerability: sensitive system and RAG configuration appeared exposed through client-server communication rather than restricted server-side. Manual verification confirmed that ordinary browser inspection allowed collection of the system prompt, model and embedding configuration, retrieval parameters, backend endpoints, API schema, document and chunk metadata, knowledge-base content, and the 1,000 most recent patient-chatbot conversations. The deployment also contradicted its privacy assurances: full conversation records, including health-related queries, were retrievable without authentication. Conclusions: Serious privacy and security failures in patient-facing RAG chatbots can be identified with standard browser tools, without specialist skills or authentication; independent review should be a prerequisite for deployment. Commercial LLMs accelerated this assessment, including under a false developer persona; assistance available to auditors is equally available to adversaries.

Public image diffusion models are now powerful enough that an attacker without the resources to train a tabular-specific generator may repurpose one off the shelf. This study tests that possibility directly. An unmodified Stable Diffusion U-Net is applied to the UCI Adult Income dataset by reshaping each row into a small single-channel pseudo-image. The architecture's inductive bias toward spatial locality makes feature placement a design variable, and several layouts are tested. However, this is only the beginning of the story, as this paper also draws two philosophical distinctions. One separates statistical from perceptual realism: whether synthetic content holds up to a machine's correlation audits or a human's sensory inspection. The other introduces synthetic evidence as a category alongside synthetic media: AI-generated material whose consumer is a machine in a closed evidentiary pipeline rather than a person in an open information system. An attacker succeeds with synthetic evidence by thinking like the machine that will receive it. And the more the attacker succeeds, the more they can induce ground truth drift: the silent reclassification of AI-generated outputs as authentic when reused in pipelines that do not interrogate their provenance.

The adoption of Internet of Things (IoT) systems at the network edge of smart architectures is increasing rapidly, intensifying the need for security mechanisms that are both adaptive and resource-efficient. In such environments, runtime defence mechanisms are no longer limited to detection alone but become a resource-constrained task of selecting mitigation actions. Security controls must be carefully selected, combined, and executed under latency, energy, and computational constraints, while preventing unsafe interactions between controls. Existing approaches predominantly rely on static rule sets and learned policies, which provide limited guarantees of feasibility, conflict safety, and execution correctness in resource-constrained edge settings. To address this limitation, we introduce ASPO, a self-adaptive multi-agent security pattern selection that integrates Large Language Model (LLM)-based reasoning with deterministic enforcement within a MAPE-K control loop. ASPO explicitly separates stochastic decision generation from execution: LLM agents propose candidate mitigation portfolios, while a deterministic optimisation core enforces closed-world action integrity, conflict-free composition, and resource feasibility at every decision epoch. We deploy ASPO on a distributed edge-gateway testbed and evaluate it across two workloads, each comprising 500 and 1000 runtime security decisions, using replayed IoT attack traffic. In addition, the results demonstrate invariant safety properties, including 100% conflict-free activation, consistent resource feasibility across workloads, and stable pattern dominance with perfect rank preservation. Importantly, deeper decision exploration reduces extreme-case execution costs, compressing tail latency and energy overheads by 21.9% and 23.1%, respectively, without increasing mean energy consumption.

Red-teaming Vision-Language Models is essential for identifying vulnerabilities where adversarial image-text inputs trigger toxic outputs. Existing approaches treat image generation as a black box, returning only terminal toxicity scores and leaving open the question of when and how toxic semantics emerge during multi-step synthesis. We introduce STARE, a hierarchical reinforcement learning framework that treats the denoising trajectory itself as the attack surface, under a direct white-box T2I and query-only black-box VLM setting. By coupling a high-level prompt editor with low-level T2I fine-tuning via Group Relative Policy Optimization (GRPO), STARE attains a 68% improvement in Attack Success Rate over state-of-the-art black-box and white-box baselines. More importantly, this trajectory-level view surfaces the Optimization-Induced Phase Alignment phenomenon: vanilla models exhibit diffuse toxicity, whereas adversarial optimization concentrates conceptual harms into early semantic phases and detail-oriented harms into late refinement. Targeted perturbations of either window selectively suppress different toxicity categories, indicating that this temporal structure is a genuine causal handle rather than a side effect of the hierarchical design. The phenomenon turns toxicity formation from a chaotic process into a small set of predictable vulnerability windows, providing both a potent attack engine and a basis for phase-aware safety mechanisms. Content warning: This paper contains examples of toxic content that may be offensive or disturbing.

As Large Language Models (LLMs) are increasingly deployed in cross-linguistic contexts, ensuring safety across diverse regulatory and cultural environments has become a critical challenge. However, existing multilingual benchmarks largely rely on general risk taxonomies and machine-translated data, limiting evaluation to predefined risk categories and providing insufficient coverage of region-specific regulatory requirements and cultural contexts. To bridge these gaps, we introduce BabelSafe, a policy-grounded multilingual safety benchmark covering 13 language settings. BabelSafe is constructed from regional regulatory sources, with risk categories and fine-grained rules extracted from jurisdiction-specific regulatory documents directly used to guide the generation of multilingual safety data. During data generation, we further incorporate region-specific cultural contexts, enabling regulation-grounded and culturally contextualized evaluation across languages. Building on BabelSafe, we develop BabelGuard, a Diffusion Large Language Model (dLLM)-based guardrail model that supports multilingual safety judgment and policy-conditioned safety assessment. BabelGuard has two variants, a lightweight 1.5B model for fast `safe/unsafe' classification and a more capable 7B model for customizable policy-conditioned safety checking with detailed explanations. We evaluate BabelGuard against 11 strong guardrail baselines on 6 existing multilingual safety benchmarks and BabelSafe, demonstrating the strong performance of BabelGuard across these evaluation settings. We hope that BabelSafe and BabelGuard can help advance the development of regulation-aware and culturally contextualized multilingual guardrail systems.

Differential Privacy (DP) has become the gold standard for protecting individual privacy in data analytics, and the shuffle-DP model has attracted significant attention from both academia and industry due to its favorable balance between privacy and utility. However, existing shuffle-DP protocols rely on a strong assumption: all users behave honestly. In real-world scenarios, adversarial users can exploit this vulnerability through poisoning attacks, compromising both privacy guarantees and the utility of analytical results. While defending against poisoning attacks in the shuffle-DP model has recently gained interest, existing solutions are limited to frequency estimation tasks. To address this issue, we propose the first general defense framework for all union-preserving queries, capable of transforming any shuffle-DP protocol into a version resilient to poisoning attacks. Beyond robust defense against poisoning attacks, our framework achieves high utility of analytical results. Compared to the original shuffle-DP protocol, it retains asymptotically equivalent error in attack-free settings and incurs only a polylogarithmic increase in error when a constant number of attackers are present. We demonstrate the generality of our framework on several common queries, including summation, frequency estimation, and range counting. Experimental results confirm that our approach effectively defends against poisoning attacks while maintaining strong utility and communication efficiency.

Trusted Execution Environments (TEEs) have emerged as a cornerstone for securing sensitive computations by providing isolated enclaves protected from untrusted software. However, their security guarantees are undermined by vulnerabilities in both the enclave code and the underlying hardware design, which can allow sensitive data to leak despite strong isolation guarantees. This paper presents KINGSGUARD, a novel TEE design that systematically monitors and controls the propagation of sensitive data within an enclave. By enforcing fine-grained data flow tracking and checks in hardware, our approach ensures that sensitive data does not leave the enclave boundary, thus bridging the gap between the idealized threat models of TEEs and their practical realizations. Additionally, to balance security with practical functionality, we introduce controlled declassification at enclave boundaries, allowing intentional release of data to the outside world. Our implementation of KINGSGUARD on a RISC-V processor has a 10.8% hardware area overhead when synthesized on FPGA and a 5.69% performance overhead.

We propose a graphical authentication scheme that follows a simple “Pick and Sort” design in which users choose visual elements and arrange them within a grid. Both the number of selected elements and the grid size are configurable, and the visual elements can be customized for specific user groups, such as children. A preliminary study with a prototype implementation indicated that the scheme is easy to learn and flexible to deploy. Although login times are longer than those of conventional authentication methods, the additional interaction may be acceptable in scenarios that are not time-critical, such as infrequent-access use cases or as a secondary authentication mechanism.

ForesightFlow is an Information Leakage Score (ILS) framework for detecting informed trading on decentralized prediction markets. For an event-resolved binary market, the score quantifies the fraction of the terminal information move priced in before the public news event. Three operational scope conditions (edge effect, non-trivial total move, anchor sensitivity) are stated as preconditions for interpretation. The score admits a Murphy-decomposition reading that connects label generation to the proper-scoring-rule literature. A pilot empirical evaluation surfaces three findings. First, a resolution-anchored proxy for the public-event timestamp does not separate event-resolved markets from a matched control population (Mann-Whitney p = 1e-6, separation reversed), demonstrating that proxy quality is itself a binding constraint. Second, the article-derived timestamp on a single high-stakes case shifts the score by 0.444 in magnitude relative to the proxy and lies on the opposite side of zero. Third, an audit of the publicly documented Polymarket insider record reveals that documented cases are systematically deadline-resolved, falling outside the original ILS scope (0 of 24 FFIC inventory markets satisfied original scope conditions). This last finding motivates a deadline-ILS extension introduced in Section 7, anchored at the public-event timestamp rather than the news timestamp, and equipped with a per-category exponential hazard baseline for the time-to-event distribution. The extension closes the gap between the methodology and the population in which insider trading has been empirically documented. An end-to-end evaluation of the extension on the 2026 U.S.-Iran conflict cluster is reported in a companion paper. We release the FFIC inventory, the resolution-typology classification of the 911,237-market corpus, and all code at github.com/ForesightFlow.

We introduce a technology to formally verify that a software system satisfies a temporal specification of functional correctness, without revealing the system itself. Our method combines a deductive approach to model checking to obtain a formal certificate of correctness for the system, with zero-knowledge proofs to convince an external verifier that the system – kept secret – complies with its specification of correctness – made public. We consider proof certificates represented as ranking functions, and introduce both an explicit-state and a symbolic scheme for model checking in zero knowledge. Our explicit-state scheme assumes systems represented as transition graphs. We use polynomial commitments to convince the verifier that the public proof certificates correspond to the secret transition relation. Our symbolic scheme assumes systems specified as linear guarded commands. We apply Farkas' lemma to obtain a witness for the validity of the ranking function and employ sigma protocols with folding to efficiently convince the verifier of the witness's existence. We built a prototype to demonstrate the practical efficacy of our two schemes on linear temporal logic verification examples. Our technology enables formal verification in domains where both the safety and the confidentiality of the system under analysis are critical.

Retrieval-augmented generation (RAG) is vulnerable to prompt injection attacks, in which an adversary inserts malicious documents containing carefully crafted injected prompts into the knowledge database. When a user issues a question targeted by the attack, the RAG system may retrieve these malicious documents, whose injected prompts mislead it into generating attacker-specified answers, thereby compromising the integrity of the RAG system. In this work, we propose CleanBase, a method to detect malicious documents within a knowledge database. Our key insight is that malicious documents crafted for the same attack-targeted questions often exhibit high semantic similarity, as attackers deliberately make them consistent to improve attack success rates. Accordingly, CleanBase constructs a similarity graph over the knowledge database, where each node represents a document and an edge connects two nodes if their semantic similarity–computed using an embedding model–exceeds a statistically determined threshold. Due to their inherent similarity, malicious documents tend to form cliques within this graph. CleanBase detects such cliques and flags the corresponding documents as malicious. We theoretically derive upper bounds on CleanBase's false positive and false negative rates and empirically validate its effectiveness. Experimental results across multiple datasets and prompt injection attacks demonstrate that CleanBase accurately detects malicious documents and effectively safeguards RAG systems. Our source code is available at https://github.com/WeifeiJin/CleanBase.

Agent skills - structured packages of instructions, scripts, and references that augment a large language model (LLM) without modifying the model itself - have moved from convenience to first-class deployment artifact. The runtime that loads them inherits the same problem package managers and operating systems have always faced: a piece of content claims a behavior; the runtime must decide whether to believe it. We argue this paper's central thesis up front: a skill is untrusted code until it is verified, and the runtime that loads it must enforce that default rather than infer trust from a signature, a clearance, or a registry of origin. Without skill verification, a human-in-the-loop (HITL) gate must fire on every irreversible call - which is operationally untenable and degrades into rubber-stamping at any non-trivial scale. With skill verification treated as a separate, gated process, HITL fires only for what is unverified, and the system becomes sustainable. We give a trust schema that includes an explicit verification level on every skill manifest; a capability gate whose HITL policy is a function of that verification level; a biconditional correctness criterion that any candidate verification procedure must satisfy on an adversarial-ensemble exercise; and a portable runtime profile with ten normative guidelines abstracted from a working open-source reference implementation. The contribution is harness- and model-agnostic; nothing here requires retraining, fine-tuning, or proprietary infrastructure.

Modern organizations increasingly rely on log data and monitoring signals to protect products against account takeovers and abuse, yet integrating security analytics into fast-moving Agile workflows remains challenging. While it is important to understand how security practices are developed and sustained within Agile, real-world case studies of such integrations remain scarce. This experience report provides insights on developer perceptions of an effort to integrate log-based fraud detection within an organization, known as the "Red Flag Project". A cross-functional team of eight members (including one author) iterated weekly to implement a proof-of-concept log-based system that alerts stakeholders when accounts exhibit suspicious activity patterns. Through semi-structured interviews, we investigate developer perceptions of log-based fraud detection integration-exploring their willingness to adopt the system, challenges encountered, and the overall impact on day-to-day development activities and security perceptions. Our findings highlight key lessons, mitigation techniques, and best practices for embedding security analytics into Agile workflows. We provide insights for practitioners and researchers seeking to incorporate security practices into modern development processes while maintaining both speed and resilience in software delivery.