Papers

Within the $f$-DP framework, we derive a tight analysis of the trade-off function for Differentially Private Stochastic Gradient Descent (DP-SGD) with subsampling based on random allocation in which each sample is independently assigned to exactly one of $M$ minibatches per epoch, each minibatch corresponding to one of the $M$ SGD rounds within a single epoch. Our analysis holds under an explicit validity condition, whose hypotheses together force $\sigma \geq \sqrt{3/\ln M}$, where $\sigma$ is the DP noise multiplier. Unlike $f$-DP analyses for Poisson subsampling, which yield non-closed implicit formulas that can be machine computed but are non-transparent, random allocation admits a tight analysis yielding transparent and interpretable closed-form bounds. For a single epoch, our concrete bounds, derived via the Berry-Esseen theorem, are tight up to constant factors. We demonstrate worked parameter settings for a single epoch ($E=1$) with a corresponding trade-off function $\geq 1-a-\delta$, that is, only $\delta$ below the ideal random guessing diagonal $1-a$. For $\delta = 1/100$ and $\sigma = 1$, roughly $M \approx 1.14\times 10^6$ rounds and $N \approx 1.14\times 10^7$ training samples suffice to achieve meaningful differential privacy. This is in contrast to recent negative results for the regime $\sigma \leq 1/\sqrt{2 \ln M}$ for which no significant DP guarantee can exist.

Large Language Models (LLMs) have revolutionized how information are collected, aggregated, and reasoned. However, this enables a novel and accessible vector of privacy intrusion: the automated and in-depth personal profiling; this engenders a chilling effect of "peepers everywhere". Existing research primarily unfolds from the training pipeline of LLM, emphasizing the exposure of Personally Identifiable Information (PII) through memorization, while privacy studies from a human-centric perspective remain underexplored. To fill this void, we empirically investigate privacy perception in the real world through the lens of human awareness and the practices of LLM-integrated platforms, revealing a significant dissonance: platforms fail to technically or policy-wise address public privacy concerns. To facilitate a systematic and quantifiable study of privacy risk, we propose the PrivacyIceberg, which categorizes real-world human privacy risks into three tiers: explicitly searched, contextually inferred, and deeply aggregated, based on the sophistication of LLM exploitation. We developed IcebergExplorer to audit privacy exposure, utilizing minimal PII as a search seed to reconstruct high-fidelity profiles, achieving over 90% factual accuracy within 10 minutes at a cost under $3, for real-world scenarios. Additionally, we identify six root causes contributing to such privacy disclosures and propose multi-stakeholder countermeasures for LLM vendors, individuals, and data publishers.

Autonomous LLM agents face a critical security risk known as workflow hijacking, where attackers subtly alter tool and skill invocations. Existing defenses rely on host-internal telemetry (such as audit logs), which can be forged if the host OS is compromised. To solve this, we introduce ClawGuard, a passive, out-of-band monitor that audits LLM-agent workflows using electromagnetic (EM) emanations. Because distinct agent skills create unique hardware usage patterns (computation, DRAM, network blocking), they emit measurable, macroscopic EM envelopes. External software-defined radios (SDRs) capture these physical signals. Using a drift-aware pipeline with 320-dimensional features, ClawGuard converts RF streams into physical evidence. Evaluated on a 7.82TB RF corpus, ClawGuard achieved an AUC of 0.9945, detecting attacks with a 100% true-positive rate and a 1.16% false-positive rate. This proves passive EM sensing is a practical, forge-resistant physical check against compromised host software.

Multi-step attacks on large language model agents may depend on opportunities distributed across sessions, such as access to target information or the availability of required tools. To combine these opportunities, an attack needs to retain its state and intermediate results, and choose actions based on current conditions. In this work, we study such attacks as cross-session attack programs. We focus on their shared control structures, including sequential execution with conditional waiting, branching and merging, looping, and condition accumulation. We represent these programs as Mealy machines and construct a sub-backdoor for each transition. We build single-session training trajectories for each sub-backdoor, combine them into a training dataset, and fine-tune the model to learn the local behaviors jointly. After a single injection of the initial trigger, the agent uses persistent memory to connect local behaviors into a complete cross-session attack program. We evaluate four instantiations of these control structures across four models in a LangChain-based agent environment. The primary instantiation achieves mean complete-program success rates of 71.7%–94.7% in LangChain. In the official OpenClaw runtime under a controlled configuration, it achieves a complete-program success rate of 85% for each of two evaluated models. These results demonstrate the feasibility of end-to-end execution of cross-session attack programs with different control structures.

The rapid emergence of generative image models has led to the development of specialized watermarking techniques, particularly in-generation methods such as seed-based embedding. However, current evaluations in this area remain largely empirical, making them heavily reliant on the specific model architectures used for generation and inversion. This prevents any clear conclusion on the performance of any method, especially regarding security, for which a rigorous definition is lacking. Against this approach, we argue that the effectiveness of a watermarking scheme should be established purely through a thorough theoretical analysis. This is enabled by decoupling the model-dependent part from the actual decision mechanism of the watermarking system. Using this decoupling, we introduce a formal evaluation framework based on security, robustness, and fidelity. This allows precise comparisons between watermarking systems through a characteristic surface representing the trade-off between these three quantities, independent of any generative model. Based on this framework, we propose SSB, a novel watermarking method that generalizes previous seed-based methods by allowing to reach any security-robustness-fidelity regime on its characteristic surface. This work opens the door to the design of modern watermarking systems with theoretical guarantees that do not necessitate any costly empirical evaluations.

The safety alignment of Large Language Models (LLMs) remains vulnerable to Harmful Fine-tuning (HFT). While existing defenses impose constraints on parameters, gradients, or internal representations, we observe that they can be effectively circumvented under persistent HFT. Our analysis traces this failure to the inherent redundancy of the high-dimensional parameter space: attackers exploit optimization trajectories that are orthogonal to defense constraints to restore harmful capabilities while deceptively adhering to safety restrictions. To address this, we propose Safety Bottleneck Regularization (SBR). SBR shifts the defensive focus from the redundant parameter space to the unembedding layer, which serves as a geometric bottleneck. By anchoring the final hidden states of harmful queries to those of the safety-aligned model, SBR enables the model to maintain safe responses even under persistent HFT. Extensive experiments confirm SBR's effectiveness, demonstrating that utilizing just a single safety anchor is sufficient to reduce the Harmful Score to $<$10 while preserving competitive performance on benign downstream tasks.

LLM agents rely on prompts to implement task-specific capabilities based on foundation LLMs, making agent prompts valuable intellectual property. However, in untrusted deployments, adversaries can copy and reuse these prompts with other proprietary LLMs, causing economic losses. To protect these prompts, we identify four key challenges: proactivity, runtime protection, usability, and non-portability that existing approaches fail to address. We present PragLocker, a prompt protection scheme that satisfies these requirements. PragLocker constructs function-preserving obfuscated prompts by anchoring semantics with code symbols and then using target-model feedback to inject noise, yielding prompts that only work on the target LLM. Experiments across multiple agent systems, datasets, and foundation LLMs show that PragLocker substantially reduces cross-LLM portability, maintains target performance, and remains robust against adaptive attackers.

GitHub Continuous Integration (CI) workflows increasingly integrate Large Language Models (LLMs) to automate review, triage, content generation, and repository maintenance. This creates a new attack surface: externally controllable workflow inputs can shape LLM prompts and outputs, which may in turn affect security decisions, repository state, or privileged execution. Although LLM security and CI security have each been studied extensively, their intersection remains underexplored. In this paper, we present the first study of LLM-induced security risks in GitHub CI workflows. We characterize the problem along the full execution chain and develop a taxonomy of high-level risk classes and concrete threat vectors. To detect such risks in practice, we design Heimdallr, a hybrid analysis framework that normalizes workflows into an LLM-Workflow Property Graph (L-WPG) and combines triggerability analysis, LLM-assisted dataflow summarization, and deterministic propagation to synthesize concrete threat-vector findings. Evaluated on 300 manually annotated unique workflows, Heimdallr achieves high accuracy on LLM-node identification (F1~=~0.994), triggerability classification (99.8%), and threat-vector detection (micro-average F1~=~0.917). As part of an ongoing detection and disclosure effort, we have so far responsibly disclosed 802 vulnerable workflow instances across 759 repositories and received 71 acknowledgments.

The New Space era has led to a rapid increase in satellites operated by independent entities in near-Earth orbit. This shift enables richer space services but also requires secure, near-real-time coordination, making efficient authentication of space assets critical for next-generation missions. Traditional ground-dependent Public Key Infrastructure (PKI) suffers from latency and operational bottlenecks that limit scalability and availability in dynamic space environments. This paper proposes architectural designs for space-based PKI that shift certificate management and validation from ground infrastructure into space, reducing reliance on ground stations while enabling interoperability and cross-entity collaboration. Two deployment schemes are introduced: a space-ground integrated PKI with in-orbit validation authorities, and a fully autonomous space-based PKI with in-space issuance and validation. We analyze deployment trade-offs in scalability, availability, security, cost, and operational complexity in multi-operator environments. A baseline latency analysis is provided to illustrate performance implications of in-orbit trust management.

Backdoor attacks can implant malicious behaviours into deep models while preserving performance on clean data, posing a serious threat to safety-critical vision systems. Although backdoor mitigation has been studied extensively for image classification, defenses for object detection remain comparatively underdeveloped. Adversarial fine-tuning is a common backdoor mitigation approach in classification, but adapting it to detection is nontrivial as classification-oriented adversarial generation does not match the detection attack space, where attacks may cause object misclassification or disappearance, and standard detection losses can dilute the repair signal across many predictions. We address these challenges through a detection-aware adversarial fine-tuning framework for mitigating object-detection backdoors when the defender has access only to a compromised detector and a small clean dataset, without knowing the attack objective. For adversarial generation that does not require knowledge of the attack objective, we introduce soft-branch minimisation, which uses a soft gate to combine objectives aligned with misclassification and disappearance attacks, together with a detection-aware classification-loss maximisation. For targeted repair, we introduce a dual-objective fine-tuning loss applied to target-matched predictions, concentrating the defensive update on predictions most relevant to the backdoor behaviour. Experiments across CNN- and Transformer-based detectors show that our approach more effectively reduces attack success while preserving true detections, compared with classification-oriented baselines, and maintains competitive clean detection performance.

Low-latency anonymity networks such as Tor remain vulnerable to infrastructure-level traffic analysis that exploits side-channel information observable from encrypted communications. We introduce NATA, a non-invasive active traffic-correlation analysis algorithm that injects distinguishable throughput patterns into traffic flows through controlled bandwidth perturbations. Unlike passive correlation methods, NATA does not require endpoint compromise, Tor-browser modification, or packet-payload decryption or modification. It can be carried out by an adversary that controls an upstream network gateway and observes traffic at adversary-controlled exit relays. To identify perturbed flows under substantial network variability, we develop BM-Net (Bandwidth Modulation Network), a selective state-space learning framework adapted for bandwidth-modulation detection. Given the limited availability of high-fidelity ground truth on real-world cross-continental Tor paths, BM-Net adopts a data-efficient learning strategy that separates self-supervised representation learning from supervised task-specific classification. It first learns reusable traffic representations through masked pre-training on serialized traffic traces, and then adapts these representations to binary perturbation detection and fine-grained modulation classification using task-specific labeled data. Through real Tor traffic measurements, BM-Net achieves a 99.65% binary detection F1 score and a 97.5% macro-F1 score for fine-grained modulation classification under our evaluated settings. In addition, tornettools-based scaled simulations are used to estimate exit-observation probability under bandwidth-weighted relay selection. These results suggest that active bandwidth perturbation can serve as an infrastructure-level side channel for traffic correlation under a clearly defined adversary model.

Agent Skills have become a practical way to extend LLM agents by packaging metadata, natural-language instructions, and executable resources into reusable capability bundles. However, this growing Skill ecosystem introduces a new compliance risk: a Skill may perform high-impact actions that fall outside the scope permitted by the user's current request, thereby violating least privilege. Existing skill detection approaches are insufficient for this problem because it is inherently task-conditioned: the same action may be legitimate under one user prompt but over-privileged under another. In this paper, we present SkillScope, a framework for fine-grained least-privilege enforcement in Agent Skills. SkillScope adopts a graph-based analysis approach that models instruction-level procedures and code-level operations as fine-grained action nodes. It extracts potential over-privilege candidates, validates them under graph-instantiated user tasks through runtime analysis, and constrains validated over-privileged actions via control-flow privilege constraining. We evaluate SkillScope through effectiveness experiments and large-scale real-world measurement. SkillScope achieves a 94.53% skill-level F1 score for over-privilege detection. In the wild, SkillScope validates 6,590 of 68,312 valid real-world Skills as exhibiting over-privileged behaviors, showing that least-privilege violations are prevalent in current Skill ecosystems. In the privilege-constraining evaluation, SkillScope reduces triggered over-privileged action-in-task instances by 88.56% while preserving legitimate task completion.

Modern LLM agents solve complex tasks by operating in iterative execution loops, where they repeatedly reason, act, and self-evaluate progress to determine when a task is complete. In this work, we show that while this self-directed loop facilitates autonomy, it also introduces a critical risk: by injecting malicious prompts into the agent's context, an adversary can distort the agent's termination judgment, making it believe the task remains incomplete and leading to unbounded computation.To understand this threat, we define and systematically characterize it as Termination Poisoning and design 10 representative attack strategies. Through a empirical study spanning 8 LLM agents and 60 tasks, we demonstrate that different LLM agents exhibit distinct behavioral signatures that determine which strategies succeed. These transferable patterns can serve as principled guidance for crafting effective attacks against previously unseen agents and tasks, enabling scalable red-teaming beyond manually designed templates. Building on these insights, we introduce LoopTrap, an automated red-teaming framework that synthesizes target-specific malicious prompts by exploiting agent behavioral tendencies. LoopTrap first constructs a behavioral profile of the target agent along four vulnerability dimensions via lightweight probing. It then performs adaptive trap synthesis, routing to the most effective strategy and selecting optimal injections via a self-scoring mechanism. Finally, successful traps are abstracted into a reusable skill library, while failed attempts are refined through self-reflection, ensuring continuous improvement. Extensive evaluation shows that LoopTrap achieves an average of 3.57$\times$ step amplification across 8 mainstream agents, with a peak of 25$\times$.

Retrieval-Augmented Generation (RAG) enables large language models (LLMs) to leverage external knowledge, but also exposes valuable RAG databases to leakage attacks. As RAG systems grow more complex and LLMs exhibit stronger instruction-following capabilities, existing studies fall short of systematically assessing RAG leakage risks. We present LeakDojo, a configurable framework for controlled evaluation of RAG leakage. Using LeakDojo, we benchmark six existing attacks across fourteen LLMs, four datasets, and diverse RAG systems. Our study reveals that (1) query generation and adversarial instructions contribute independently to leakage, with overall leakage well approximated by their product; (2) stronger instruction-following capability correlates with higher leakage risk; and (3) improvements in RAG faithfulness can introduce increased leakage risk. These findings provide actionable insights for understanding and mitigating RAG leakage in practice. Our codebase is available at https://github.com/yeasen-z/LeakDojo.

LLMs are increasingly explored for malware analysis; however, current LLM-based malware attribution remains limited by unsupported indicators and insufficient code-level grounding for identifying malicious and vulnerable code segments. To address these limitations, this research introduces LCC-LLM, a code-centric benchmark dataset and evidence-grounded framework for malware attribution and multi-task static malware analysis. The proposed LCCD dataset contains approximately 34K PE samples processed through a large-scale reverse-engineering pipeline and represented using decompiled C code, assembly code, CFG/FCG artifacts, hexadecimal data, PE metadata, suspicious API evidence, and structural features. Beyond dataset construction, LCC-LLM integrates LangGraph-orchestrated static analysis with multi-source cybersecurity knowledge to support evidence-grounded malware reasoning. The framework employs a seven-layer retrieval-augmented generation pipeline, CoVe for IoC validation, and a multi-dimensional quality gate to improve factual reliability and analyst-oriented decision support. Curriculum-ordered instruction data is used to fine-tune DeepSeek-R1-Distill-Qwen-14B and Qwen3-Coder-30B-A3B using QLoRA. Evaluation across 43 malware-analysis task types achieves an average semantic similarity of 0.634, with the highest task-level performance in structured report generation, IoC extraction, vulnerability assessment, malware configuration extraction, and malware class detection. In a real-world case study using MalwareBazaar samples, the grounded pipeline achieves a 10/10 structured analysis pass rate, producing CFG/FCG evidence, MITRE ATT&CK mappings, detection guidance, and analyst-ready reports. These results show that code-centric representations, retrieval grounding, and verification-guided reasoning improve the reliability and operational usefulness of LLM-assisted malware attribution.

Image steganography is widely used to protect user privacy and enable covert communication. However, it can also be abused by the adversary as a covert channel to bypass content moderation, disseminate harmful semantics, and even hide malicious instructions in images to elicit dangerous outputs from large models, posing a practical security risk that continues to evolve. To address the lack of a unified and systematic evaluation framework, we propose SADBench, a systematic benchmark that assesses the adversary's ability to inject harmful secrets via steganography and the defender's ability to detect such threats through steganalysis. Crucially, SADBench comprises $4$ core tasks, namely steganography attack capability evaluation, steganalysis defense capability evaluation, efficiency evaluation, and transferability evaluation. It evaluates both image-payload and text-payload steganography across diverse cover distributions, utilizing harmful visual semantics and toxic instructions to simulate malicious attacks. Across a broad set of attacks and detectors, SADBench reveals that (i) INN and autoencoder-based methods demonstrate superior stability compared to other architectures, (ii) in-domain detection is near-perfect and cheaper than generation, (iii) a critical asymmetry exists in transferability where attacks robustly generalize to new distributions while detectors fail to adapt, and (iv) real-world threats persist on social media, where payloads either survive minimal compression or effectively adapt to aggressive compression via simulated training. Overall, SADBench establishes a systematic, reproducible, and extensible framework to quantify risks, paving the way for measurable and security-driven advancements in steganography defense.

Most production ERC-4337 Paymasters rely on Process-Oriented Abstraction (POA): a centralized off-chain server signs each sponsorship request, acting as a potential censorship bottleneck. We propose Asset-Oriented Abstraction (AOA), encapsulating payment capability in a persistent, user-owned on-chain asset – the Gas Card – rather than an off-chain signing process. Following the Design Science Research (DSR) methodology, we implement SuperPaymaster on Optimism Mainnet, anchoring sponsorship validity in on-chain Soulbound Token state and deterministic policy rules, removing the off-chain signer as a validity gate. We evaluate gas costs via single-UserOp ERC-20 transfers on Optimism Mainnet (n = 50 per system). In pure L2 execution gas (txGasUsed; actualGasUsed = txGasUsed + PVG), SuperPaymaster (167,830) is lower than both evaluated POA baselines: Alchemy Gas Manager (205,951) and Pimlico ERC-20 paymaster (328,937). It still pays a ~32,000-gas on-chain verification overhead versus Alchemy, but reduces gas by 49% versus Pimlico by replacing on-chain token liquidation with an internal balance update. In total billed gas, SuperPaymaster (286,818) exceeds Alchemy (257,299) due to higher bundler PVG overhead, not paymaster architecture. Code structural analysis and on-chain Mainnet evidence confirm that sponsorship validity requires no off-chain signing server: validatePaymasterUserOp reads only on-chain state. These findings suggest that AOA can mitigate the usability-decentralization-efficiency trade-offs in gas payment.

This paper introduces the $\alpha$-Wasserstein mechanism for achieving Rényi Pufferfish Privacy using Laplace and Gaussian noise. By leveraging Hölder's inequality, we demonstrate that the scale parameter of the Laplace mechanism can be calibrated via an upper bound on the $W_\alpha$ metric to satisfy $(\alpha, \epsilon)$-Rényi Pufferfish Privacy for $\alpha \in (1, \infty]$. We show that at the limit $\alpha = \infty$, this framework recovers the established $W_\infty$ mechanism for $\epsilon$-pufferfish privacy. This result is subsequently extended to the exponential mechanism. Furthermore, we propose a $W_\alpha$ mechanism for Gaussian noise for $\alpha \in (1, \infty)$, demonstrating that it generalizes existing results within the Rényi Differential Privacy framework. Experimental evaluations reveal that our $\alpha$-Wasserstein mechanism significantly reduces noise power compared to the conventional $W_\infty$-based approach, with the Gaussian mechanism providing superior utility over the Laplace mechanism. Notably, the mechanisms derived in this work achieve exact $(\alpha, \epsilon)$-Rényi Pufferfish Privacy without requiring additional relaxations, such as $\delta$-approximations.

Recent advances in foundation models have transformed LLMs from passive conversational systems into autonomous agents capable of reasoning and tool execution. While these capabilities unlock substantial practical value, they also introduce new security risks, as adversaries can manipulate agents into performing harmful actions in real-world environments. Existing defense strategies mitigate such threats but frequently struggle to balance safety and utility, resulting in over-refusal of benign user requests. To mitigate this trade-off, we propose SafeHarbor, a novel framework designed to establish precise decision boundaries for LLM agents. Unlike static guidelines, SafeHarbor extracts context-aware defense rules through enhanced adversarial generation. We design a local hierarchical memory system for dynamic rule injection, offering a training-free, efficient, and plug-and-play solution. Furthermore, we introduce an information entropy-based self-evolution mechanism that continuously optimizes the memory structure through dynamic node splitting and merging. Extensive experiments demonstrate that SafeHarbor achieves state-of-the-art performance on both ambiguous benign tasks and explicit malicious attacks, notably attaining a peak benign utility of 63.6% on GPT-4o while maintaining a robust refusal rate exceeding 93% against harmful requests. The source code is publicly available at https://github.com/ljj-cyber/SafeHarbor.

Privacy-preserving action recognition (PPAR) enables machines to understand human activities in videos without revealing sensitive visual content. Among the various strategies for PPAR, encryption-based methods achieve strong privacy protection while maintaining high recognition performance. However, these methods lead to a catastrophic decrease in recognition performance and visual quality when the encrypted videos are compressed. That is, the previous methods are not compression-friendly. To address these issues, in this paper, we propose the first compression-friendly encryption method for PPAR, called CFE-PPAR. In CFE-PPAR, videos encrypted with secret keys can be directly recognized by a video transformer, which uses parameters transformed by the same keys as those used for video encryption. In experiments, it is verified that CFE-PPAR outperforms previous methods on the UCF101 and HMDB51 datasets under Motion-JPEG and H.264 compression.

Federated learning (FL) is attractive for cloud-edge intrusion detection because it enables collaborative training over distributed telemetry without centralizing raw logs. In production security analytics pipelines, however, only a subset of clients participates in each round, and heterogeneous bandwidth, stragglers, and dropouts can cause the server to rely on stale client information. This paper studies client participation as a timeliness-aware systems problem using Age of Information (AoI). We compare three lightweight policies for federated intrusion detection: AoI-first, utility-first, and a hybrid AoI+utility rule with a tunable trade-off parameter. Across a CIC-IDS2017 DDoS/PortScan mini subset, NSL-KDD, ToN-IoT, and a synthetic drift benchmark under clean, poisoning, and poisoning-plus-robust-aggregation settings, AoI-aware selection reduces average AoI by about 39–41% and peak AoI by about 70% relative to random sampling while keeping the per-round communication budget fixed. The hybrid policy usually preserves Macro-F1/AUC and provides an interpretable knob for balancing freshness, detection quality, and robustness, although it is not uniformly Pareto-dominant once false positive rate is included. Robustness is evaluated by combining AoI-guided selection with trimmed-mean aggregation under label-flip poisoning; the selection policy itself is not intended as a standalone Byzantine defense. The main practical message is that cloud-edge, privacy-preserving intrusion analytics can improve timeliness through a lightweight scheduling layer without changing the underlying FL participation budget.

Retrieval-Augmented Generation (RAG) systems are vulnerable to knowledge base poisoning, yet existing attacks have been evaluated almost exclusively against vanilla retrieve-then-generate pipelines. Architectures designed to handle conflicting retrieved information - multi-agent debate, agentic retrieval, recursive language models - remain untested against adversarially optimized contradictions. We evaluate four RAG architectures (vanilla RAG, agentic RAG, MADAM-RAG, and Recursive Language Models) under controlled single-document (N=1) poisoning on 921 Natural Questions QA pairs, comparing a clean baseline, naive injection, and CorruptRAG-AK - an adversarial attack whose meta-epistemic framing targets credibility assessment. Architecture is a high-impact variable in adversarial robustness: under CorruptRAG-AK, attack success rates range from 81.9% (vanilla) to 24.4% (RLM) - a spread of nearly 58 percentage points across architectures with comparable clean accuracy (~92%). Decomposing this gap, once the poisoned document is retrieved, adversarial framing - not retrieval optimization - drives the majority of CorruptRAG-AK's advantage for three of four architectures, localizing the cross-architecture vulnerability at the content-reasoning stage. Our MADAM-RAG reimplementation shows the highest apparent contradiction detection rate, though our LLM judge over-identifies this behavior (~48.5% precision), so reported rates are upper bounds. Regardless of detection, MADAM-RAG cannot resolve contradictions reliably, producing a 41.4% non-answer rate even on clean inputs - though implementation divergences from the original may contribute. We introduce a seven-category behavioral taxonomy capturing contradiction detection, hedging, and failure modes beyond binary accuracy. Code, data, and analysis notebooks are publicly available.

Hidden malicious intent in multi-turn dialogue poses a growing threat to deployed large language models (LLMs). Rather than exposing a harmful objective in a single prompt, attackers can distribute their intent across multiple benign-looking turns, making defense a problem not only of whether a dialogue is harmful, but also of when intervention becomes necessary. Existing trace-level labeling approaches provide only coarse safety signals and do not identify this intervention boundary, making it difficult to distinguish timely intervention from premature refusal or a block that comes too late. This work introduces turn-level harm-enabling supervision for multi-turn defense. We define the earliest harm-enabling turn as the first point at which delivering a candidate response would make the accumulated interaction sufficient to enable harmful action. To instantiate this supervision at scale, we construct the Multi-Turn Intent Dataset (MTID), which contains adaptive attack rollouts, matched benign hard negatives, and annotations of this boundary. Using MTID, we train TurnGate, a response-aware monitor that learns when to intervene, and further optimize its policy through multi-turn reinforcement learning. Experiments show that turn-level boundary supervision improves intervention localization, while reinforcement learning further improves the safety–utility trade-off. TurnGate outperforms existing guardrails and multi-turn monitoring baselines, and generalizes across risk domains, attacker pipelines, and target models. Our code is available at https://github.com/Graph-COM/TurnGate.

An emerging blockchain protocol design pattern leverages the asymmetry between the computational effort in performing versus verifying tasks. For example, cryptographic validity proofs (e.g., SNARKS) require the prover to expend significant effort demonstrating the correctness of their claim, while the verifiers benefit from extremely easy validation. The operationalization of this paradigm requires efficiently soliciting the performance of expensive tasks in pseudonymous, adversarial environments. We formalize this as a mechanism design question. The protocol balances the economic cost of a liveness fault, where the work is not completed, with the payments required to incentivize specific behavior from candidate suppliers. We show that the loss of the optimal protocol scales logarithmically in the cost of a liveness fault, scaled up by the adversarial fraction of the network. Further, we find that the optimal equilibria have an intuitive structure, allowing us to provide concrete advice to practitioners. Specifically, in many regimes, the optimum designates a single, random node as the primary worker and a committee as a fallback, which is reminiscent of leader-based consensus mechanisms. We also characterize the asymptotic regimes where having negative payments (i.e., slashing in blockchain parlance) is especially helpful.

Effective security logging is crucial for the timely and accurate detection of cyber threats; however, the relative effectiveness of various industry-standard logging frameworks remains understudied. This paper addresses this critical gap by presenting the first systematic evaluation of modern security logging standards utilizing a novel methodology built upon the automated Security Exploit Telemetry Collection (SETC) framework. SETC systematically generates reproducible exploit scenarios in containerized environments, collecting rich telemetry across multiple logging standards, including CIM (Common Information Model), OCSF (Open Cybersecurity Schema Framework), and ECS (Elastic Common Schema). The detection efficacy of each logging standard is quantified by measuring telemetry completeness and exploit detectability across standardized logs through detailed experiments involving 50 diverse remote code execution vulnerabilities. The resulting findings identify critical gaps and reveal significant differences in logging standards' abilities to capture key attack indicators. Our contributions include a novel evaluation methodology that enables scalable and reproducible analysis of exploit telemetry, as well as new findings that provide clear, evidence-based guidance for security practitioners to make informed decisions about adopting logging standards.

Deploying large language models in financial-services and agentic settings requires safety classifiers that simultaneously handle prompt injection, regulatory compliance, and general harm, a combination no existing open guardrail addresses in a single inference pass. Semalith v1.4 is a 184M-parameter DeBERTa-v3-base classifier performing simultaneous three-axis safety classification including prompt injection, general harm, and financial-services regulatory compliance, in a single forward pass. Its 22-class head (BENIGN, nine prompt-injection sub-types, general-harm, eleven BFSI labels) is trained with a 4-class auxiliary super-category head under jointly weighted loss, on a 76,204-row corpus mined from 49 public sources with SHA-1 deduplication against every held-out evaluation set, with 21 of 22 benchmarks at zero contamination (max 0.22%). Against Llama-Guard-3-8B on 22 held-out benchmarks, Semalith v1.4 wins every prompt-injection evaluation (7/7) and 11 of 18 benchmarks overall at 44x fewer parameters, with FPR = 0.000 on 208 benign agentic prompts vs 0.063 for Llama-Guard-3-8B. On general-harm benchmarks (WildGuardMix, HEx-PHI, HarmBench), Llama-Guard-3 leads; this complementary split is documented in Section 4. Six measured weak spots are disclosed in Section 6. Deployment guidance: v1.3 is recommended for conversational moderation deployments (ToxicChat F1 0.624); v1.4 is recommended when BFSI label coverage or zero-FPR on benign agentic prompts is the priority.

The rapid diffusion of agentic AI has created a new coverage problem for commercial insurance: some AI-mediated losses are now affirmatively insured, some create silent-AI exposure under legacy cyber, technology errors-and-omissions (E&O), directors-and-officers (D&O), employment practices liability (EPLI), crime, and media policies, and others are being actively excluded. This paper maps that emerging boundary by coding 55 AI threat classes against 26 insurance products, endorsements, and exclusion regimes using public carrier materials and OWASP/MITRE threat catalogs. We identify a four-tier insurability frontier: affirmatively insured perils, silent-AI exposures, actively excluded perils, and perils outside conventional private insurance structures. Our coding measures publicly claimed positioning rather than executed contract wording; the headline statistics describe what carriers publicly state about coverage, not what would be paid in any specific claim. Three patterns emerge. First, affirmative AI coverage is beginning to differentiate by primary risk emphasis: public materials often position Munich Re around model performance and drift, Armilla and parts of the Lloyd's market around hallucination and broader AI liability, Tokio Marine Kiln and CFC around IP and technology E&O concerns, Apollo ibott around emerging autonomous system liability, and Coalition around deepfake and AI-enabled cyber response. Second, legacy lines retain silent-AI exposure where AI is an instrumentality rather than the legal cause of loss. Third, foundation model concentration is the clearest genuinely novel insurability frontier because upstream model failure can correlate losses across many cedents at once; the relevant market design question is which insurability constraint each candidate structure relaxes, not merely which systemic risk template exists.

Agentic AI systems can plan, call tools, inspect code, interact with web applications, and coordinate multi-step workflows. These same capabilities change the economics of cyber offense. The central near-term risk is not that every low-skill criminal immediately becomes a frontier exploit researcher; it is that agentic AI compresses the attack lifecycle by lowering the cost of reconnaissance, phishing, credential abuse, vulnerability triage, exploit adaptation, and post-compromise decision support. This paper synthesizes current public evidence from national cybersecurity agencies, industry threat reports, agent security guidance, and research on LLM agents cyber capabilities. It introduces a Three Channel Agentic Cyber Risk Model and an Agentic Attack Compression Model, uses the 2026 Linux kernel Copy Fail incident as a case study for foothold-to-root acceleration, and develops a 2026 to 2028 forecast for large enterprises and the German and European Mittelstand. The paper concludes with a prioritized defense roadmap. Organizations should treat agentic AI security as an immediate operational problem: identity, phishing resistant authentication, patch velocity, CI/CD and Linux/container hardening, agent governance, telemetry, and recovery readiness must be strengthened now.

Age verification before accessing restricted content is critical to protecting minors from exposure to harmful material such as pornography, gambling, violence, hateful speech, and substance purchases like alcohol and tobacco. Currently, the absence of reliable age-checking mechanisms allows children extensive access to such adult content, posing significant risks to their worldview and mental development. While regulatory efforts like the European Union's Digital Services Act promote using Digital Wallets or Age Verification Apps, relying solely on government-based solutions raises concerns about data sensitivity and privacy risks. Effective age verification must therefore be trustworthy, user-friendly, privacy-preserving, and offer flexible assurance levels. We analyze currently implemented (UK or Australia) and proposed (UE) solutions from different angles, pointing out the weaknesses and threats, and come up with an alternative. Our proposal addresses these challenges by leveraging open standards - such as Privacy Pass and Privacy Access Tokens - and cryptographic techniques to enable secure, privacy-conscious age verification without requiring specialized software installation. This approach empowers users to select trusted providers from multiple options, reducing the risk of data breaches and ensuring a safer digital environment for minors.

Large language models (LLMs) are increasingly being integrated into web browsers to create agentic browsing systems that execute actions on behalf of the user. Prior work considering the security of agentic browsers focuses exclusively on indirect prompt-injection attacks. However, by failing to consider traditional web attacks, previous agentic browser threat models have a blind spot to web social engineering attacks originally designed to trick humans. In this paper, we propose the first web-focused threat model for agentic browsers and use it to derive a taxonomy of 20 attacks across both the web and LLM space, and implement 18 of the attacks. Our threat model extends the original See$\rightarrow$Act browser agent model to account for all components of a browser, and frames the agent as a confused deputy unable to distinguish task steps from traditional web attacks. We show that 10 web threats can reemerge often in amplified forms once an agent can be influenced by untrusted page content. We further conduct a generalizability study on 14 of the 20 attacks, showing that our attacks reproduce across 4 major LLM models spanning multiple vendors. We show that agentic browsers exhibit five major failure modes when facing traditional and LLM web threats, demonstrating the need to rearchitect agentic browsers before they are ready for the current web.