Papers

Recent multi-bit watermarking methods for large language models (LLMs) prioritize capacity over reliability, often conflating decoding with detection. Our analysis reveals that existing ECC-based extractors suffer from catastrophic false positive rates (FPR), and applying rejection thresholds merely collapses detection sensitivity (TPR) to random guessing. To resolve this structural limitation, we propose BREW (Block-wise Reliable Embedding for Watermarking), a framework shifting the paradigm to designated verification. BREW employs a two-stage mechanism: (i) blind message estimation via independent block voting, followed by (ii) window-shifting verification that rigorously validates the payload against local edits. Experiments demonstrate that BREW achieves a TPR of 0.965 with an FPR of 0.02 under 10% synonym substitution, demonstrating that the high-FPR issue is not an inherent trade-off of multi-bit watermarking, but a solvable structural flaw of prior decoding-centric designs. Our framework is model-agnostic and theoretically grounded, providing a scalable solution for reliable forensic deployment.

An agent skill is a configuration package that equips an LLM-driven agent with a concrete capability, such as reading email, executing shell commands, or signing blockchain transactions. Each skill is a hybrid artifact-a structured half declares executable interfaces, while a prose half dictates when and how those interfaces fire-and the prose is reinterpreted probabilistically on every invocation. Conventional static analyzers parse the structured half but ignore the prose; LLM-based tools read the prose but cannot reproducibly prove that a tainted input reaches a high-impact sink. We present Semia, a static auditor for agent skills. Semia lifts each skill into the Skill Description Language (SDL), a Datalog fact base that captures LLM-triggered actions, prose-defined conditions, and human-in-the-loop checkpoints. Synthesizing a fact base that is both structurally sound and semantically faithful to the original prose is the central challenge; we address it with Constraint-Guided Representation Synthesis (CGRS), a propose-verify-evaluate loop that refines LLM candidates until convergence. Security properties (e.g., indirect injection, secret leakage, confused deputies, unguarded sinks, etc.) over an agent skill can then be reduced to Datalog reachability queries. We evaluate Semia on 13,728 real-world skills from public marketplaces. Semia renders all of them auditable and finds that more than half carry at least one critical semantic risk. On a stratified sample of 541 expert-labeled skills, Semia achieves 97.7% recall and an F1 of 90.6%, substantially outperforming signature-based scanners and LLM baselines.

Traditionally, machine learning methods for PE malware detection have relied on static features like byte histograms, string information, and PE header contents. One barrier to incorporating dynamic analysis features has been the semi-structured nature of sandbox behavior reports. We show that, using the latest generation of large language models with reasoning, it is possible to efficiently process these behavior reports and utilize them as part of a malware detection pipeline. Specifically, we leverage LLMs to generate behavior-based malware detection rules based on a small training set of labeled malware. We find that these detection rules, derived from behavioral features, are much more robust to concept drift than standard static-feature methods, while maintaining practical false positive rates. Finally, we introduce Trident, a system which combines a classic decision tree model over static features, our behavior-based detection rules, and direct LLM analysis of sandbox reports through majority voting. Trident outperforms standard methods using static features, outperforms behavior-based rules alone, and is as resilient to concept drift as active learning methods without requiring retraining.

Conformance checking, one of the main process mining operations, aims to identify discrepancies between a process model and an event log. The model represents the expected behaviour, whereas the event log represents the actual process behaviour as captured in information systems records. Traditionally, the process model and the event log are both accessible to the business analyst performing the conformance checking. However, in some contexts, it is necessary to keep either the model or the log private to protect critical or sensitive information. In this paper, we propose a secure approach to conformance checking based on string processing algorithms and homomorphic encryption, where the process model and event log ar not visible to either the model's or event log's owner. The proposed technique is based on alignments, a well-known formalism used for conformance checking. An evaluation is performed using a synthetic and a real-world event log, showing that conformance checking can be securely computed at the expense of high memory and processing requirements.

AI-powered edge computing security is moving Intelligent Transportation Systems (ITS) from passive, rule-based protections to proactive, smart, zero-touch, self-sufficient safeguards that neutralize threats in milliseconds. As transportation becomes more connected with edge computing, massive IoT, and advanced 5G for vehicle-to-everything (V2X) connectivity, AI at the edge computing nodes plays a crucial role in protecting against sophisticated threats, enabling URLLC (ultra-low-latency communications) for smart transport, and enhancing infrastructure capabilities and safety. This research applies edge computing to improve latency, bandwidth efficiency, and service responsiveness by moving processing closer to devices, gateways, and users. However, this shift also expands the cyberattack surface because edge nodes are distributed, heterogeneous, and often resource-constrained. The paper proposes a trust-aware federated hybrid intrusion detection framework in which a random forest, a decision tree, and a linear SVM network learn complementary traffic representations at each edge site, while a server performs trust-aware aggregation of local model updates.

As language model safeguards become more robust, attackers are pushed toward developing increasingly complex jailbreaks. Prior work has found that this complexity imposes a "jailbreak tax" that degrades the target model's task performance. We show that this tax scales inversely with model capability and that the most advanced jailbreaks effectively yield no reduction in model capabilities. Evaluating 28 jailbreaks on five benchmarks across Claude models ranging in capability from Haiku 4.5 to Opus 4.6, we find Haiku 4.5 loses an average of 33.1% on benchmark performance when jailbroken, while Opus 4.6 at max thinking effort loses only 7.7%. We also observe that across all models, reasoning-heavy tasks display considerably more degradation than knowledge-recall tasks. Finally, Boundary Point Jailbreaking, currently the strongest jailbreak against deployed classifiers, achieves near-perfect classifier evasion with near-zero degradation across safeguarded models. We recommend that safety cases for frontier models should not rely on a meaningful capability degradation from jailbreaks.

Safety-aligned large language models rely on RLHF and instruction tuning to refuse harmful requests, yet the internal mechanisms implementing safety behavior remain poorly understood. We introduce the Attention Redistribution Attack (ARA), a white-box adversarial attack that identifies safety-critical attention heads and crafts nonsemantic adversarial tokens that redirect attention away from safety-relevant positions. Unlike prior jailbreak methods operating at the semantic or output-logit level, ARA targets the geometry of softmax attention on the probability simplex using Gumbel-softmax optimization over targeted heads. Across LLaMA-3-8B-Instruct, Mistral-7B-Instruct-v0.1, and Gemma-2-9B-it, ARA bypasses safety alignment with as few as 5 tokens and 500 optimization steps, achieving 36% ASR on Mistral-7B and 30% on LLaMA-3 against 200 HarmBench prompts, while Gemma-2 remains at 1%. Our principal mechanistic finding is a dissociation between ablation and redistribution: zeroing out the top-ranked safety heads produces at most 1 flip among 39 to 50 baseline refusals, while ARA targeting the corresponding safety-heavy layers flips 72/200 prompts on Mistral-7B and 60/200 on LLaMA-3. This suggests that safety is not localized in these heads as removable components, but emerges from the attention routing they perform. Removing a head allows compensation through the residual stream, while redirecting its attention propagates a corrupted signal downstream.

Mobile remote identity verification (RIdV) systems are exposed to attacks that manipulate or replace the facial video stream, including presentation attacks, real-time deepfakes, and video injection. Recent European requirements, including ETSI TS 119 461 and CEN/TS 18099, motivate evidence channels beyond camera-based presentation-attack detection. This paper studies whether passive motion traces recorded during selfie capture provide auxiliary evidence for spoof screening and user verification. We introduce CanSelfie, a dataset of 375 bona fide multi-sensor sequences collected at 50 Hz from 30 participants using a commercial mobile RIdV application, together with stationary, handheld, and temporally shifted attack-proxy scenarios. We benchmark 7 multivariate time-series classifiers and 8 whole-series anomaly detectors across sensor configurations and temporal windows. For spoof screening, accelerometer-only ROCKAD obtains 0.00% FRR and 43.8% FAR, while QUANT+3-NN obtains the lowest FAR of 32.0% at 2.37% FRR; both reject all stationary proxies. For same-device and same-session user verification, WEASEL+MUSE reaches 1.07% EER using 9 channels. Raw acceleration is the most informative modality, and closed-set accuracy alone does not imply good verification because threshold calibration depends on score distributions. The findings support selfie-capture motion as a low-friction auxiliary signal, while leaving cross-device, cross-session, and real injection-attack evaluation as necessary next steps. Code and data available at: https://github.com/Ergzar/Selfie-Motion-AD-TSC.

Phishing webpages are continuously polluting the Web. Plenty of countermeasures have been proposed and the most advanced techniques leverage machine-learning methods that infer whether a webpage is benign or not by inspecting its visual representation. Yet, despite the demonstrated effectiveness of such detection methods, this class of defenses is, by design, susceptible to a kind of subtle-but-cheap timing-based attacks which – worryingly, and perhaps surprisingly – have never been investigated so far. Such an oversight questions the overall reliability of these defenses in the wild. First, we show that timing-based evasion attacks have not been accounted for by prior work on visual phishing websites detectors. Then, we elucidate the intrinsic vulnerability of these detectors: they can be bypassed by delaying the rendering of webpage elements. Practically, these detectors must compute the visual similarity between a target webpage and a known legitimate one. This requires taking a "snapshot" of the target webpage before the similarity computation. Attackers can deliberately delay the rendering of key elements, such as the logo, so that these elements appear fully only after the snapshot has been taken. This simple tactic misleads the visual-similarity module, leading the system to incorrectly classify the phishing page as benign. We empirically show that state-of-the-art detectors can be completely defeated (detection rate dropping from 100% to 0%) by employing easy-to-apply problem-space techniques such as curtain effects. We also carry out a user study, evaluating the effectiveness of these attacks against real humans, and find that end users are unable to reliably identify our "perturbations" (p<.05). Finally, we propose mitigations, including a browser-extension that, without making any call to remote services, warns users that they may have landed on a phishing webpage.

Wide exploration on robocall surveillance research is hindered due to limited access to public datasets, due to privacy concerns. In this work, we first curate Robo-SAr, a synthetic robocall dataset designed for robocall surveillance research. Robo-SAr comprises of ~200 unwanted and ~1200 legitimate synthetic robocall samples across three realistic adversarial axes: psycholinguistics-manipulated transcripts, emotion-eliciting speech, and cloned voices. We further propose RoboKA, a Kolmogorov-Arnold Network (KAN)-based multimodal fusion framework designed to model structured nonlinear interactions between acoustic and linguistic cues that characterize diverse adversarial robocall strategies. RoboKA first leverages cross-modal contrastive learning to align latent modality representations and feeds the resulting embeddings to a KAN-projection head for final classification. We benchmark RoboKA against strong unimodal and multimodal baselines in both in-domain and out-of-domain setups, finding RoboKA to surpass all baselines in terms of recall and F1-score.

Online signature verification (OSV) requires distinguishing skilled forgeries from genuine samples under high intra-class variability and with very few enrollment samples. Existing deep learning methods operate directly on raw temporal sequences, restricting them to 1D architectures and preventing the use of pretrained 2D vision backbones. We bridge this gap with GAFSV-Net, which represents each signature as a six-channel asymmetric Gramian Angular Field image: three kinematic channels (pen speed, pressure derivative, direction angle) are each encoded into complementary GASF and GADF matrices that capture pairwise temporal co-occurrence and directional transition structure respectively. A dual-branch ConvNeXt-Tiny encoder processes GASF and GADF independently, with bidirectional cross-attention enabling each branch to query discriminative patterns from the other before metric-space projection. Training uses semi-hard triplet loss with skilled-forgery hard-negative injection; verification is performed via cosine similarity against a small enrollment prototype. We evaluate on DeepSignDB and BiosecurID, outperforming all sequence-based baselines trained under identical objectives, demonstrating that the representational gain of 2D temporal encoding is consistent and independent of training procedure, with ablations characterising each design choice's contribution.

Agentic security systems increasingly combine LLM planners with tools that can discover, validate, and report vulnerabilities. This creates an asymmetric control problem: the system should retain strong offensive capability inside an authorized engagement, while the same capabilities must be denied outside scope. Existing guardrails provide useful policy controls, but they do not make this boundary a first-class formal contract over observable effects. We introduce alignment contracts, a framework for specifying and enforcing behavioral constraints over observable effect traces. A contract defines scope, allowed and forbidden effects, resource budgets, and disclosure policies. We give the language finite-trace semantics, characterize satisfaction as a safety property with finite violation witnesses, develop refinement and one-way composition rules for modular contract engineering, and show that admissibility checking is decidable. We instantiate the framework for web-focused agentic security workflows and show how the same structure extends to other effect profiles. Under an explicit Effect Observability Assumption, where all $\SigmaEff$-effects are mediated, the soundness theorem quantifies over the agent model and gives guarantees for mediated $\SigmaEff$-effects, including enforcement soundness for monitor-realized traces. We also state an assumption-lifted adaptation result and formalize limits through undecidability transfer and observability-boundary theorems. A Lean 4 artifact checks the formal core theorems used by the paper.

Software Bills of Materials (SBOMs) are increasingly mandated by regulators, yet existing sharing mechanisms impose a binary choice between full disclosure and full opacity. This exposes software suppliers to attacks that can be deduced from the SBOM only, such as the presence of a vulnerable dependency. Conversely, software consumers can be fooled by software suppliers who modify or misrepresent published SBOMs. We present zkSBOM, a privacy-preserving SBOM sharing mechanism designed to address these threats. zkSBOM uses zero-knowledge sets to cryptographically commit to the components within an SBOM. Software consumers can query for known vulnerabilities and receive a cryptographic proof confirming whether the artifact described by the SBOM is affected, without revealing any additional SBOM content. We conduct a security analysis of zkSBOM by quantifying expected leakage from inclusion and exclusion proofs. We demonstrate real-world feasibility by applying it to realistic scenarios and evaluating its operation requirements. Our evaluation demonstrates that zkSBOM is a strong, secure, and privacy-preserving mechanism for SBOM sharing, protecting software suppliers and software consumers from one another.

We present XekRung, a frontier large language model for cybersecurity, designed to provide comprehensive security capabilities. To achieve this, we develop diverse data synthesis pipelines tailored to the cybersecurity domain, enabling the scalable construction of high-quality training data and providing a strong foundation for cybersecurity knowledge and understanding. Building on this foundation, we establish a complete training pipeline spanning continued pre-training (CPT), supervised fine-tuning (SFT), and reinforcement learning (RL) to further extend the model's capabilities. We further introduce a multi-dimensional evaluation system to guide the iterative improvement of both domain-specific and general-purpose abilities. Extensive experiments demonstrate that XekRung achieves state-of-the-art performance on cybersecurity-specific benchmarks among models of the same scale, while maintaining strong performance on general benchmarks.

Agentic payment systems extend delegated action to financial transfers, but scaling them on stablecoin rails in regulated settings requires safeguards that remain effective when humans are not continuously in the loop. We present a compliance-aware architecture that combines x402-style, signature-based payment authorisation and relayed execution with programmable compliance embedded as an on-chain guardrail via a policy wrapper and policy manager coordinating modular checks. By enforcing compliance at the point of execution, rather than as a separate off-chain workflow, the approach preserves low-friction settlement when conditions are satisfied, records transaction-linked on-chain attestations, and supports structured resolution when requirements are pending.

Integrity of audit logs produced by Internet of Things (IoT) devices is a prerequisite for post-incident forensics, regulatory compliance, and operational accountability. While blockchain-backed logging infrastructures can satisfy this requirement, they introduce consensus overhead, network dependencies, and deployment complexity that are often prohibitive at the IoT edge. This paper presents a lightweight and evaluated integrity verification pipeline that combines Merkle-tree commitments with resource-aware adaptive chunking to provide tamper evidence without relying on distributed ledger technologies. The proposed pipeline operates in three stages: (i) resource-aware batch ingestion via adaptive chunk sizing, (ii) Merkle-tree construction with O(logn) inclusion proof generation, and (iii) deterministic single-entry verification against a trusted root anchor. We further report an implementation audit that identified and corrected two evaluation defects: a double-counting bug in tampering metrics and a redundant full-tree reconstruction during batch appends. Using the corrected implementation, five-run benchmarks on synthetic IoT log datasets demonstrate throughput exceeding 130,000 logs/s for 100,000 records. The system achieves per-entry verification latency of approximately 22 ms, proof generation latency of 22 ms, an average proof size of 1,006 bytes, and peak memory usage below 5 MB. Tampering detection achieves perfect precision, recall, and F1-score (1.0) across corruption ratios ranging from 1% to 50%.

Long-context large language models (LLMs)-for example, Gemini-3.1-Pro and Qwen-3.5-are widely used to empower many real-world applications, such as retrieval-augmented generation, autonomous agents, and AI assistants. However, security remains a major concern for their widespread deployment, with threats such as prompt injection and knowledge corruption. To quantify the security risks faced by LLMs under these threats, the research community has developed heuristic-based and optimization-based red-teaming methods. Optimization-based methods generally produce stronger attacks than heuristic attacks and thus provide a more rigorous assessment of LLM security risks. However, they are often resource-intensive, requiring significant computation and GPU memory, especially for long context scenarios. The resource-intensive nature poses a major obstacle for the community (especially academic researchers) to systematically evaluate the security risks of long-context LLMs and assess the effectiveness of defense strategies at scale. In this work, we propose FlashRT, the first framework to improve the efficiency (in terms of both computation and memory) for optimization-based prompt injection and knowledge corruption attacks under long-context LLMs. Through extensive evaluations, we find that FlashRT consistently delivers a 2x-7x speedup (e.g., reducing runtime from one hour to less than ten minutes) and a 2x-4x reduction in GPU memory consumption (e.g., reducing from 264.1 GB to 65.7 GB GPU memory for a 32K token context) compared to state-of-the-art baseline nanoGCG. FlashRT can be broadly applied to black-box optimization methods, such as TAP and AutoDAN. We hope FlashRT can serve as a red-teaming tool to enable systematic evaluation of long-context LLM security. The code is available at: https://github.com/Wang-Yanting/FlashRT

Multi-turn prompt injection follows a known attack path – trust-building, pivoting, escalation but text-level defenses miss covert attacks where individual turns appear benign. We show this attack path leaves an activation-level signature in the model's residual stream: each phase shift moves the activation, producing a total path length far exceeding benign conversations. We call this adversarial restlessness. Five scalar trajectory features capturing this signal lift conversation-level detection from 76.2% to 93.8% on synthetic held-out data. The signal replicates across four model families (24B-70B); probes are model-specific and do not transfer across architectures. Generalization is source-dependent: leave-one-source-out evaluation shows each of synthetic, LMSYS-Chat-1M, and SafeDialBench captures distinct attack distributions, with detection on real-world LMSYS reaching 47-71% when its distribution is represented in training. Combined three-source training achieves 89.4% detection at 2.4% false positive rate on a held-out mixed set. We further show that three-phase turn-level labels(benign/pivoting/adversarial) unique to our synthetic dataset are essential: binary conversation-level labels produce 50-59% false positives. These results establish adversarial restlessness as a reliable activation-level signal and characterize the data requirements for practical deployment.

Decompositional jailbreaks pose a critical threat to large language models (LLMs) by allowing adversaries to fragment a malicious objective into a sequence of individually benign queries that collectively reconstruct prohibited content. In real-world deployments, LLMs face a continuous, untraceable stream of fully anonymized and arbitrarily interleaved requests, infiltrated by covertly distributed adversarial queries. Under this rigorous threat model, state-of-the-art defensive strategies exhibit fundamental limitations. In the absence of trustworthy user metadata, they are incapable of tracking global historical contexts, while their deployment of generative models for real-time monitoring introduces computationally prohibitive overhead. To address this, we present TwinGate, a stateful dual-encoder defense framework. TwinGate employs Asymmetric Contrastive Learning (ACL) to cluster semantically disparate but intent-matched malicious fragments in a shared latent space, while a parallel frozen encoder suppresses false positives arising from benign topical overlap. Each request requires only a single lightweight forward pass, enabling the defense to execute in parallel with the target model's prefill phase at negligible latency overhead. To evaluate our approach and advance future research, we construct a comprehensive dataset of over 3.62 million instructions spanning 8,600 distinct malicious intents. Evaluated on this large-scale corpus under a strictly causal protocol, TwinGate achieves high malicious intent recall at a remarkably low false positive rate while remaining highly robust against adaptive attacks. Furthermore, our proposal substantially outperforms stateful and stateless baselines, delivering superior throughput and reduced latency.

System auditing on Android faces two problems. First, existing syscall tracers lose events under load, silently overwriting entries faster than a user space reader can drain them. Second, security-relevant application behavior is mediated through Binder, Android's kernel IPC mechanism, and is therefore hidden from the syscall layer. The Binder parcels that the kernel does see carry no method names or typed arguments, a disconnect between low-level events and high-level behavior known as the semantic gap. Existing approaches address the semantic gap either by modifying the Android platform, making them difficult to adjust to OS updates, or by instrumenting the traced application in user space, which sophisticated adversaries can evade by bypassing the instrumented framework APIs. We present WOOTdroid, a design and prototype for on-device tracing on stock Android that addresses both problems without OS modification or application instrumentation. WDSys, an eBPF port of eAudit-style syscall auditing, runs on current Android with at most 3.6% Geekbench overhead and traces 33% more syscalls than ftrace. WDBind captures Binder parcels in the kernel and decodes them out-of-process against a framework signature table extracted via Java reflection. We demonstrate WOOTdroid on Pixel 9 devices running Android 16 with an end-to-end case study reconstructing ten security-relevant Binder transactions.

Mixture-of-Experts (MoE) architectures in Large Language Models (LLMs) have significantly reduced inference costs through sparse activation. However, this sparse activation paradigm also introduces new safety challenges. Since only a subset of experts is engaged for each input, model behavior becomes coupled to routing decisions, yielding a difficult-to-control mechanism that can vary across safety-relevant scenarios. At the same time, adapting model behavior through full fine-tuning or retraining is costly, especially when developers need to rapidly configure the same model for different safety objectives. We present MASCing (MoE Activation Steering Configuration), the first framework that enables flexible reconfiguration of MoE behavior across diverse safety scenarios without retraining. MASCing uses an LSTM-based surrogate model to capture cross-layer routing dependencies and map routing logits to downstream behaviors. It then optimizes a steering matrix to identify behavior-relevant expert circuits and, at inference time, applies steering masks to the routing gates to override expert selection. This enables targeted enhancement or suppression of specific behaviors while preserving general language utility. To demonstrate its reconfigurability, we apply MASCing to two different safety-related objectives and observe consistent gains with negligible overhead across seven open-source MoE models. For multi-turn jailbreak defense, it improves the average defense success rate from 52.5% to 83.9%, with gains of up to 89.2%. For adult-content generation, MASCing enables models to comply with such requests that would otherwise be refused, increasing the average generation success rate from 52.6% to 82.0%, with gains of up to 93.0%. These results establish MASCing as a practical, lightweight, and flexible framework for scenario-specific safety reconfiguration in MoE models.

The rapid proliferation of image generation models and other artificial intelligence (AI) systems has intensified concerns regarding data privacy and user consent. As the availability of public datasets declines, major technology companies increasingly rely on proprietary or private user data for model training, raising ethical and legal challenges when users request the deletion of their data after it has influenced a trained model. Machine unlearning seeks to address this issue by enabling the removal of specific data from models without complete retraining. This study investigates a modified SISA (Sharded, Isolated, Sliced, and Aggregated) framework designed to achieve class-level unlearning in Convolutional Neural Network (CNN) architectures. The proposed framework incorporates a reinforced replay mechanism and a gating network to enhance selective forgetting efficiency. Experimental evaluations across multiple image datasets and CNN configurations demonstrate that the modified SISA approach enables effective class unlearning while preserving model performance and reducing retraining overhead. The findings highlight the potential of SISA-based unlearning for deployment in privacy-sensitive AI applications. The implementation is publicly available at https://github.com/SiamFS/ sisa-class-unlearning.

Lattice reduction smooths the Gram-Schmidt profile, and we use majorization to describe the local swap mechanism behind that smoothing. In this language, each non-degenerate Lovász swap acts as a T-transform on the log-norm profile. As a consequence, every strictly Schur-convex measure of profile spread decreases at such a swap. Two structural consequences follow. First, the worst-case GSA envelope admits a variational interpretation. It is the unique minimum-variance profile compatible with the Lovász gap geometry, so its slope is determined by the LLL parameter alone. Second, the realized swap trajectory satisfies an exact telescoping identity for variance dissipation. The same viewpoint also helps organize deep-insertion heuristics. It suggests a thermal family of Schur-convex scoring rules, motivates adaptive selection within that family, and leads to two concrete selectors: Thermal-Adaptive, which reduces operation counts relative to SS-GG on flat profiles in our benchmarks while recovering SS-GG on $q$-ary inputs, and Geodesic Deep-LLL, which reduces equivalent-swap counts on structured lattices in our benchmarks at higher wall-clock cost.

How code representation format shapes false positive behaviour in cross-language LLM vulnerability detection remains poorly understood. We systematically vary training intensity and code representation format, comparing raw source text with pruned Abstract Syntax Trees at both training time and inference time, across two 8B-parameter LLMs (Qwen3-8B and Llama 3.1-8B-Instruct) fine-tuned on C/C++ data from the NIST Juliet Test Suite (v1.3) and evaluated on Java (OWASP Benchmark v1.2) and Python (BenchmarkPython v0.1). Cross-language FPR reflects the joint effect of training-time and inference-time representation, not either alone. Text fine-tuning drives FPR upward monotonically (Qwen3-8B: 0.763 zero-shot, 0.866 pilot, 1.000 full-scale) while F1 remains stable (0.637-0.688), masking the collapse. We argue surface-cue memorisation is the primary mechanism: text fine-tuning encodes C/C++-specific API names and syntactic idioms as vulnerability triggers that fire indiscriminately on target-language code. A cross-representation probe, applying text-trained weights to AST-encoded input without retraining, isolates this: Qwen3-8B FPR drops from 0.866 to 0.583, and 37.2% of false positives revert to true negatives under AST input alone. Direct AST fine-tuning does not preserve the benefit (FPR at least 0.970), as flat linearisation introduces structural surface cues of its own. The pattern replicates across both model families. On BenchmarkPython the AST probe yields FPR=0.554, within 2.9 percentage points of the Java result, despite maximal surface-syntax differences, substantially weakening a domain-shift explanation. These findings motivate a pre-deployment consistency gate, running alerts through both text and AST paths, as a retraining-free filter for false-positive-sensitive settings, at the cost of reduced recall.

Renewed attention to the identity of Bitcoin's pseudonymous creator has revived an old worry: that the roughly 1.148 million BTC mined by Satoshi and never moved represent a major tail risk for bitcoin. This paper argues that the worry is overstated. The mechanical downside of selling the position is bounded well below the feared collapse, and the outcomes most consistent with sixteen years of observed behavior are not bearish for bitcoin's effective supply. We analyze the position in two ways. First, we model the case of a purely financial holder. Multiple sale scenarios, checked against both a square-root-law estimate and the historical record of large sales, suggest that bitcoin's current market liquidity could absorb a patient multi-year sale with a cumulative price impact centered around 10 to 13 percent relative to a no-sale case. The same arithmetic also links the downside from a surprise sale to the upside from a confirmed burn: both are bounded by the same effective-supply adjustment, so the doom case and the burn-rally case cannot both be large. Second, we consider the preferences implied by the sixteen-year record. Ideological restraint, privacy, already having enough, and preserving the myth all point toward further dormancy, permanent loss of access, or a deliberate burn. A sale or an act of sabotage remains possible, but the record supports it less strongly. Under both approaches, the mechanical bear case is bounded, and the likeliest outcomes are neutral to mildly positive for bitcoin's effective supply. The argument does not rule out transient overshoot or leverage-driven amplification; it bounds the durable repricing the coins themselves can cause.

The hubness problem, in which hub embeddings are close to many unrelated examples, occurs often in high-dimensional embedding spaces and may pose a practical threat for purposes such as information retrieval and automatic evaluation metrics. In particular, since cross-modal similarity between text and images cannot be calculated by direct comparisons, such as string matching, cross-modal encoders that project different modalities into a shared space are helpful for various cross-modal applications, and thus, the existence of hubs may pose practical threats. To reveal the vulnerabilities of cross-modal encoders, we propose a method for identifying the hub embedding and its corresponding hub text. Experiments on image captioning evaluation in MSCOCO and nocaps along with image-to-text retrieval tasks in MSCOCO and Flickr30k showed that our method can identify a single hub text that unreasonably achieves comparable or higher similarity scores than human-written reference captions in many images, thereby revealing the vulnerabilities in cross-modal encoders.

Regulation requires generated text to be marked and detectable, but symmetric-key watermarks leave detection to the key-holding provider. We study first-party provenance audit (FPA), where an organization examines sensitive content in its custody. Such an audit should not disclose the audited text and should provide a verifiable result and checkable record. We present VOW, which uses a verifiable oblivious pseudorandom function to evaluate the detector's keyed component without revealing its inputs to the detection service, while the provider retains the watermarking key. A signed setup record published before use fixes the public key and detector specification, and a proof binds returned evaluations to that key. The auditor computes the statistic and decision locally, so the service may refuse but cannot make an incorrect result pass verification. Given the text, transcript, and setup record, a reviewer can reconstruct the prescribed result without the watermarking key or provider interaction. With the detector configured for a nominal false-positive rate of $10^{-5}$, VOW detects 99.4% of 200-token watermarked texts with competitive text quality and practical overhead. Under explicit conditions, our oracle-assisted text construction produces detector-positive text for all three Green-Red detectors evaluated. Paraphrasing substantially weakens every scheme evaluated, limiting watermark-based FPA to relatively intact content.

Formal verification provides rigorous guarantees for cryptographic security, yet extracting formalizable security goals from natural-language protocol documents remains largely manual. We introduce SecGoal, a dedicated expert-annotated dataset and benchmark for extracting formalizable security goal statements from protocol documents, covering 15 widely deployed protocols, together with AIFG, a schema- and flow-conditioned framework for structured formal security property generation. Our evaluation shows that frontier and large LLMs achieve high property recall but low extraction precision because they often fail to distinguish formalizable security goals from non-goal protocol content. In contrast, SecGoal fine-tuning makes smaller open-source LLMs substantially more selective extractors of formalizable security goals. On the held-out test protocols, Gemma2-9B-FT improves extraction precision from 24.0% to 66.6% and reaches 97.6% property recall, outperforming larger prompted LLMs and encoder baselines. In a controlled setting, AIFG shows that concise goal inputs can support high-recall structured property generation, while expert-vetted extracted inputs reveal over-generation as the main remaining bottleneck. Together, SecGoal and AIFG provide a dataset, benchmark, and framework for specification-grounded security goal extraction and property generation.

Boolean circuits form the foundational computational substrate of symmetric cryptography, yet the exploration of their architectural design space has remained largely confined to a handful of canonical paradigms - SPN, Feistel networks, and their immediate variants. This paper takes a deliberately broader perspective by formalizing the design space of cryptographic Boolean systems through six independent binary structural constraints: Stratification, Acyclicity, Regularity, Interleaving, Homogeneity, and Locality. These constraints generate a hypercube of $2^6 = 64$ distinct architectural classes defined over Synchronous Boolean Networks, a general model that subsumes both acyclic combinational circuits and recurrent synchronous systems. We systematically evaluate all 64 classes against three generic cryptanalytic fitness objectives - differential, linear and algebraic resistance - using a five-stage methodology centered on Formal Concept Analysis. The results reveal that the best Boolean networks are governed by the identification of sparse, mutually compatible combinations of constraints - a fundamentally epistatic problem that classical cryptography has barely addressed.

As web browsers increasingly restrict client-side tracking, the web tracking ecosystem is shifting from client-side to server-side tracking (SST). In SST, the browser sends tracking requests to an intermediate endpoint, which then forwards them to the tracker's endpoint, eliminating direct client-to-tracker requests. As a result, existing tracking protections that block requests to known tracker endpoints are rendered ineffective. In this paper, we investigate server-side implementation of Google Analytics, the most widely deployed third-party tracking service on the web today. We also present SST-Guard, a multi-modal, browser-based system for detecting and blocking server-side Google Analytics (sGA). Our key insight is that even when the tracker's endpoints change, sGA must necessarily still collect and share the same semantic information as client-side Google Analytics (e.g., identifiers, event metadata). Therefore, rather than detecting requests to known Google Analytics endpoints, SST-Guard aims to detect underlying artifacts of collection and sharing of these semantic values to any arbitrary endpoint. Operationalizing this insight is challenging because real-world sGA deployments commonly customize endpoints and obfuscate URLs/payloads. SST-Guard addresses this challenge using a value-template approach that employs regular expressions to match semantic value patterns across multiple modalities: network requests, cookies, and the window object. We validate SST-Guard on Tranco top-10k websites, detecting 4.02% (403) sGA domains with over 93% accuracy across three modalities, with network request classifier demonstrating the highest accuracy (99.8%). By deploying SST-Guard in the wild, we find 4.21% (6,314) of Tranco top-150k websites using sGA.