Papers

Self-evolving skill (SES) systems distill agent trajectories into persistent skills, allowing untrusted experience to become trusted instruction. We introduce PoisonedEvolution, a trajectory-poisoning attack on this promotion process. Our skill-visible black-box attacker can inspect a target skill and contribute bounded evidence, but cannot observe private pools or evolution logic or edit the skill bank. Artifact poisoning requires Inclusion, Evolution Attribution, and Realization. Attribution is the distinctive bottleneck: the target behavior must appear causally useful, recurrent, and generalizable before promotion. We evaluate four representative security-effect families using inert canary specifications. At 10% attacker support, across six mainstream LLM evolvers in SkillClaw, PoisonedEvolution embeds target behaviors in 546/600 trials (91.0% SER). On the structurally different Trace2Skill pipeline at the same ratio, it embeds target behaviors in 369/600 trials (61.5% SER), demonstrating transfer across evolution architectures. In a representative controlled study, three consistent attacker records suffice in a 30-record batch, whereas a single record is much weaker. Ablations identify recurring support, causal framing, and domain-aligned encoding as the main determinants of success. These findings expose evidence promotion as a security boundary for self-evolving agents.

Deep learning models have become integral to security applications due to their ability to model complex relationships in data and detect sophisticated threats. However, their complexity makes it difficult to understand how predictions are generated, posing significant challenges for interpretability, particularly in security applications where transparency is critical. Existing explanation methods, such as visual explanation techniques and post-hoc approaches, suffer from several limitations: reduced faithfulness due to local approximation errors, instability caused by reliance on randomness, and computational inefficiency that hinders real-time usage. To address these issues, we introduce XSec, a self-explainable deep architecture developed for security applications. During training, XSec uses a novel mask-based approach to extract informative sub-features from the data and learns prototypes, representative patterns that characterize each class. XSec then leverages the prototypes in a dedicated similarity layer at test time to compute similarity scores and generates interpretable explanations without the need for post-hoc analysis. We evaluate XSec across five diverse security scenarios, demonstrating its ability to achieve an average classification accuracy of 97.33% with minimal performance compromise. XSec produces deterministic explanations for a fixed trained model and input and substantially reduces explanation latency compared with approximation-based and perturbation-based post-hoc methods. Through this effort, we extend the applicability of self-explainable AI to security applications, bridging the gap between deep learning performance and the need for explainability in critical scenarios.

Modern vehicles rely on the Controller Area Network (CAN) bus, whose design prioritizes low cost and real-time performance but provides no message authentication or encryption. An attacker with physical or remote access can therefore inject arbitrary frames, making intrusion detection an important defense-in-depth mechanism. Most published CAN intrusion detection systems rely on presence-based features, such as novel arbitration IDs, frozen payload bytes, or anomalous DLC values. These features perform well on public datasets containing easily separable attacks but fail when attackers reuse legitimate arbitration IDs. We present per-ID behavioral residualization, a CAN-specific representation that extracts fourteen temporal, protocol, and payload features from sliding windows and residualizes them against each arbitration ID's normal baseline. Our central claim is that this representation, rather than any individual detector, drives the performance gains. Across six unsupervised detectors and two datasets, residualization improves mean F1 in the majority of evaluations (21/24 on HCRL and 30/36 on ROAD across five seeds). On the more realistic ROAD dataset, where attacks reuse legitimate IDs, the representation achieves recall >= 0.99 with high ROC-AUC on targeted signal-manipulation attacks. Two limitations are explicitly quantified: novel-ID flooding (HCRL DoS, F1 = 0.02) and cross-ID fuzzing (ROAD, F1 = 0.27), defining the measured coverage boundary of the proposed representation.

Smart-home assistants increasingly use multimodal large language models (MLLMs) that perceive video and audio directly. This raises a safety question specific to the home: can the agent tell a genuine user command from ambient or externally-sourced content, television speech, on-screen text, or an overheard conversation, that merely looks like a command? We introduce PromptShield-Home, a pilot benchmark of realistic smart-home scenarios spanning addressee ambiguity, screen/audio injection, health-monitor false triggers, mixed occupancy, and a legitimate-command floor, and use it to compare three abstraction layers: traditional detectors (L0), a single MLLM agent (L1; vision, vision+ASR, and audio-visual), and multi-agent mediation (L2; voting, role specialists, cross-model arbitration). Because the label distribution is skewed toward inaction, aggregate accuracy is misleading, a constant always-block predictor scores 82%, so we report unsafe-execution and safe-completion rates separately. The two paradigms fail in opposite ways: detectors act on everything, while every MLLM configuration over-refuses, completing almost no genuine command and missing a true fall in every case. Crucially, their correct sets are disjoint: an oracle that always picks the right layer reaches 94.1%, against 76.5% for the best single layer. We report this as an upper bound, not a system - no router is implemented - and argue that home-agent safety is best served by learned routing and sensor fusion, not by replacing detectors with an MLLM.

Connected aftermarket devices extend vehicle diagnostics, repair workflows, and over-the-air software maintenance beyond original equipment manufacturer boundaries, yet their heterogeneous ownership and long service life complicate conventional perimeter security. This paper develops Zero Trust Federated Learning for Connected Aftermarket Devices (ZT FL CADE), an edge-learning architecture that combines device-level access control, privacy-preserving federated learning, and adversarial validation for over-the-air update and predictive maintenance decisions. The evaluation uses a single synthetic dataset of 144,000 telemetry windows from 240 devices, 12 vendor domains, and 180 days of operation. Features include bus entropy, update latency, attestation age, signature retries, environmental signals, fault-code rates, packet loss, drift, mileage, and trust score, with targets for maintenance risk, update intrusion, and access action. ZT FL CADE trains local temporal models, aggregates privacy-bounded updates, scores each device against behavioral and update integrity evidence, and routes update requests to allow, challenge, or quarantine actions. Synthetic experiments improve maintenance risk F1 from 0.837 for Fed Avg to 0.883, improve intrusion F1 from 0.856 to 0.897, and preserve 0.842 intrusion recall when 20 percent of selected clients are adversarial. Mean access-decision latency remains 44 MS, below the 100 MS operational budget used in the simulation. The results do not establish field validation, but they indicate that zero-trust policy enforcement and federated learning can be evaluated jointly rather than as separate aftermarket security controls. Index Terms Zero trust architecture, federated learning, connected aftermarket devices, over-the-air updates, adversarial machine learning, edge artificial intelligence, predictive maintenance, automotive cybersecurity

Watermarking traces the provenance of text produced by large language models by embedding statistically detectable signals during decoding. Existing schemes fall into logits-based, sampling-based, entropy-aware, and adaptive-strength families, yet all of them place watermark signals according to local token statistics. In the open-ended text-generation settings evaluated in this work, local statistics may provide insufficient guidance for placing robust watermark signals. We introduce WorldMark, a plug-and-play interface that uses World Knowledge Memory (WKM) to organize semantic and episodic knowledge in a memory graph, converts the retrieved knowledge into a token-level knowledge saliency score, and adjusts the strength of a host watermark through Asymmetric Knowledge Modulation (AKM). WorldMark requires no backbone retraining and introduces no additional detector-side model or parameter. On the primary C4 evaluation, the complete WorldMark interface improves clean and attacked detection across three adaptive-strength host variants while slightly reducing perplexity. Additional pilot experiments on C4 and OpenGen show that direct memory conditioning transfers across multiple watermark families but can be unstable without saliency-aware modulation. WorldMark requires no additional detector-side model or parameter and introduces negligible overhead under the primary protocol.

The systematic and excessive data collection practices of tech companies have rendered online privacy both a necessity and a sought-after commodity. However, while the privacy risks of the web, mobile, and IoT ecosystems have been extensively examined, desktop environments have been largely overlooked. As desktop apps continue to be widely used, they remain a critical yet understudied dimension of user privacy. In this paper, we address this gap by presenting the first, to our knowledge, comprehensive study of the mechanisms designed to regulate and disclose data collection and sharing practices in the macOS ecosystem. We adopt an app-development-centric view, and shed light on the interactions between the various macOS mechanisms that mediate apps' data access. Driven by our findings, we develop NutriScan, an analysis framework that incorporates both static and dynamic analysis techniques to create a consolidated view of macOS apps' data practices and disclosures. We use our system to dynamically analyze 1K macOS apps, and find that 85% of them access user-data APIs without disclosing it. 49.7% also exfiltrate data to advertising entities and hosting providers, 12.5% of which do so without a corresponding disclosure. We find that desktop apps are being leveraged by online trackers to enrich user profiles and device fingerprints, thus shedding new light on the true scope of the online tracking ecosystem. Our analysis reveals how the macOS app ecosystem is comprised of disjoint mechanisms with divergent data abstractions, thus increasing complexity for developers while also facilitating undisclosed privacy-invasive practices. Accordingly, we propose a series of mitigations that aim to both streamline the data disclosure process for developers and improve Apple's app vetting process.

Bitcoin's consensus depends not only on protocol rules but on the emergent behavior of a heterogeneous network of nodes with divergent economic stakes. Whether a contentious soft fork resolves cleanly or fractures into a persistent chain split is difficult to predict analytically but can be explored through controlled experimentation. This paper applies Scenario Discovery, an ensemble-simulation methodology using the Patient Rule Induction Method (PRIM), to identify the configuration regions that produce contentious fork outcomes. Using Warnet to run real bitcoind nodes across 1,330 valid scenarios spanning economic weight, mining-pool commitment, pool ideology, hashrate, and difficulty-retarget regime, we identify the parameter thresholds separating clean resolution from contested split. We show that economic weight distribution across a partition, not hashrate majority, is the primary determinant of resolution under Bitcoin's operational retarget interval and moderate price divergence. Three findings follow: an economic-support floor (0.45-0.50) and override ceiling (0.78-0.82) bound the contested space, with an Economic Self-Sustaining Point (0.74) between them; a pool-commitment "flip-point" (0.214 of committed hashrate) at which committing the largest pool to the upgrading chain paradoxically reverses the outcome; and outcomes resolve on two independent layers, hashrate and economic adoption, governed by different parameters. Individual user nodes show no detectable influence on fork outcomes under the modeled economic weightings. We close with three monitoring questions, answerable from public data, that translate these thresholds into operational guidance.

The remarkable instruction-following ability of modern LLMs has enabled their practical use as the minds of agents that can autonomously complete increasingly complex tasks. Therein, however, also lies their vulnerability to attacks which embed malicious instructions in text, common variants of which are known as indirect prompt injection (IPI). A fundamental task in addressing this vulnerability is successful segmentation of a given text into benign and malicious sentences (if any). While a number of approaches for this task have been proposed, no detector combines query-relative detection at the segment level, and none are hardened against adaptive evasion attacks realizable in agentic executions. We address the former limitation by developing an approach for malicious sentence classification that is both context- and query-aware. Next, to harden the resulting classifier against evasion, we present two adversarial training methods. The first is directly adapted feature-space adversarial training (AT) in which evasions are approximated using projected-gradient-based optimization in the embedding space. The second simulates realizable evasion attacks in the AT loop through LLM-based paraphrasing. Crucially, we parametrize both AT variants to facilitate a smooth tradeoff between utility and attack robustness. In extensive experiments using indirect prompt injection benchmarks we show that the proposed approach outperforms state-of-the-art IPI defense baselines under static attacks, while in the case of adaptive attacks, our AT variants provide significantly higher utility, lower attack success rate, and often both. Finally, we show that the best AT parameters can depend intimately on the particular application domain. Consequently, domain-dependent tuning of malicious text detectors is likely necessary in practice. Our code is publicly available at https://github.com/tavia-liu/CAD.

This paper presents a reverse engineering analysis of a firmware update for a commercial vehicle Brake ECU. We analyze the updater executable to extract firmware and perform differential binary analysis of the S12X architecture firmware images. Our research identifies specific changes in the recall that address undocumented vulnerabilities in legacy protocol processing. We demonstrate that the patched functionality contained critical flaws. The findings confirm the safety recall remediation was also a security patch.

Unicity is a novel blockchain infrastructure for enabling users to execute off-chain peer-to-peer token transactions while preventing parallel states of tokens (double-spending) with minimal blockchain complexity and storage. A key component of the infrastructure is the Aggregation Layer responsible for storing information about the spent states of tokens and providing compact cryptographic proofs of no double-spending for the users without making any compromises in trust. Aggregation Layer is a layer 2 style service that holds an append-only key-value repository that periodically certifies its state using a traditional blockchain that we call the Consensus Layer. Every time while certifying a changed state the Aggregation Layer provides a cryptographic proof to the Consensus Layer about the append-only consistency of the key-value store. We use the Radix Sparse Merkle Trees (RSMTs) to create the cryptographic digest r_i of the store in every round i and authentication paths of special type in order to prove that the next digest r_{i+1} was obtained while only adding some key-value pairs (k,v) to the repository. The proof verification code is implemented as an Algebraic Intermediate Representation (AIR) circuit on top of the Plonky3 STARK toolkit. Our implementation uses no trusted setup, achieves throughput of 10,000 insertions per second and and millisecond range verification time on a single consumer-class CPU.

Modern cloud applications often comprise thousands of microservices whose interactions form complex request paths. Traditional inter-service access control restricts individual service-to-service requests, but fails to prevent multi-hop attacks, where each hop appears legitimate yet the overall path violates security intent. This gap leaves systems exposed to unauthorized access and data exfiltration. In this paper, we present eMicro, a path-aware defense system for microservices that prevents such attacks while remaining efficient and deployable. eMicro enforces real-time multi-hop access control through three key techniques: (1) history-based access control extended to capture service invocation sequences; (2) security policies encoded as efficient deterministic finite automaton (DFA), supporting constant-time lookups and compact label propagation; (3) eBPF-based in-kernel request tracing for transparent, low-overhead enforcement without code changes. Evaluations on DeathStarBench and production cloud traces from Uber, Alibaba, and ByteDance, covering 12 million request workflows and thousands of services, demonstrate the scalability of eMicro. eMicro performs policy checks in 1 microsecond, stores 50 million policies in only 100 MB, and reduces propagation overhead by 90% with negligible runtime impact. These results show that eMicro delivers scalable and efficient protection against multi-hop attacks, making it practical for deployment in large-scale microservice environments.

Retail investors lack access to the kind of personalized, tax-aware portfolio management that institutional clients take for granted – existing robo-advisors use static, rule-based allocation, and institutional-grade systems require account minimums and technology stacks unavailable to individual investors. We present a fully built, integration-tested application that closes this gap: a FastAPI backend and web dashboard that let a user describe an investment goal in plain language (e.g. "I want steady growth but need to sell some shares next month for a down payment"), routes that goal to one of six investment mandates, and produces a live, broker-integrated portfolio recommendation from athree-phase reinforcement learning system – a self-supervised cross-asset encoder, a Mixture-of-Experts (MoE) allocation policy with a learned intent router, and a lightweight LoRA adapter that personalizes recommendations from an individual's revealed brokerage behavior without retraining the shared model. The system is functionally complete and integration-tested end-to-end against a live brokerage API (Alpaca, paper-trading mode), including multi-user authentication, a trust first preview-before-apply confirmation flow, daily email digests, and an auditable action-integrity chain, but has not yet been opened to real end-users; we report this honestly as an emerging, pre-deployment application with a concrete path to full deployment, alongside 14-day walk-forward backtests (bootstrapped confidence intervals included) as preliminary, pre-deployment validation rather than production performance. We also report several practical engineering lessons – silently-inactive integration paths, hanging third-party API calls, and the value of end-to-end empirical verification over trusting checkpoint metadata – that we believe generalize to other applied RL systems built on external, live data sources.

Public-safety agencies and critical infrastructure operators rely on trunked land-mobile radio (LMR) systems, based on P25, TETRA, and TETRAPOL. These systems are expected to protect not just the content of a communication but the fact of it. Yet LMR standards leave a stark gap between confidentiality of content and of communication: underneath an encrypted traffic plane, their signaling plane is almost entirely in the clear. We probe the depth and impact of adversarial inference from this exposed signaling. Prior security analyses of these systems have concentrated on the content plane—recovering encryption keys or capturing accidental cleartext. We show that comparably sensitive information can be inferred from passively observed signaling even if the content encryption were perfect. In particular, we show that across the trunked LMR standards, a passive, receive-only software-defined radio (SDR) observer can recover operationally sensitive network topology and geography details, unit presence, mobility across cells and groups, organizational structure, as well as operational security details such as special key domains and key-epoch rotation. This signaling-plane inference reaches far beyond the observer's direct area of reception, turning local sniffing into nationwide network mapping capabilities that degrade or defeat LMR standards' identity obfuscation through timing and association. In the case of TETRAPOL, we demonstrate how inference and tracking of such signaling metadata and a standards-level confidentiality failure in emergency call handling enable unencrypted voice extraction. Finally, we discuss potential countermeasures and mitigations, including specific recommendations for protecting inter-cell, base station and subscriber identities.

This full research paper presents CyberBridge, a framework that automatically maps cybersecurity job descrip- tions to professional role profiles. As the cybersecurity landscape evolves rapidly, it is difficult for academic programs to align curricula with the competencies expected in practice. CyberBridge addresses this gap by decomposing a given vacancy description into its constituent Knowledge, Skill, and Task (KST) statements, embedding them using a sentence- BERT model, and matching them to the most semantically similar workforce profiles. A key contribution of our approach is its interpretability. Rather than functioning as a black box, CyberBridge enables users to trace recommendations back to the specific competencies driving each match, providing a human- readable justification for the resulting mappings. CyberBridge supports three primary use cases; (1) job recom- mendation, providing students with vacancies and closely match- ing professional roles based on their completed curriculum, (2) market analysis, enabling educators and curriculum developers to analyze which roles are currently in demand, and (3) curriculum planning, assessing which course program best prepares students for in-demand cybersecurity roles. As such, CyberBridge can be used by educational institutions as a practical tool for career guidance and evidence-based curriculum development in cybersecurity education.

Autonomous coding agents are increasingly embedded in enterprise software workflows with delegated authority over connected systems. Central to this architecture is the agent skills interface: folders of instructions and scripts that agents load dynamically to specialize their behavior. This interface also widens the attack surface, letting malicious shell commands hide within natural-language skill files. We make three contributions. First, an adversarial skill-synthesis method using six LLMs across four families to transform 471 real-world shell commands into benign-appearing skills, released as a benchmark of 2,826 skills mapped to 11 MITRE ATT&CK tactics. Second, a reproducible evaluation pipeline coupling run stratification, evidence anchoring, a refusal veto, and a deterministic declared-intent override with a three-judge LLM-as-a-judge panel, validated against a blind human gold standard (Cohen's kappa = 0.85). Third, a large-scale characterization of two enterprise-grade agents across 5,629 completed runs. Gemini CLI is exploited in 95.5-96.1% of runs and Qwen Code in 71.6-74.0% (raw majority vote to declared-intent-corrected estimate, both within the human gold standard), nearly invariant to the generating model. Explicit safety recognition occurs in only 1.99% of runs. Enterprises must assess and mitigate skill-interface risk before adopting coding agents. Our code and dataset are available at https://github.com/awsm-research/AgentJailbreak

Vision Transformers (ViTs) increasingly rely on input-adaptive inference, such as token pruning and early halting, to meet energy and latency budgets. This survey examines a recent class of adversarial efficiency degradation attacks that target these mechanisms to increase computation without necessarily degrading accuracy. We unify and compare two representative attacks, SlowFormer (a universal adversarial patch) and DeSparsify (per-image perturbations), across three popular token-pruning frameworks: A-ViT, ATS, and AdaViT. We standardize reporting using GFLOPs, accuracy loss, and an Attack Success (AS) metric that measures how much of the model's compute savings the attack takes away. Understanding these attacks is crucial for designing countermeasures that not only mitigate risk but also remain lightweight, since deployment often occurs in low-power settings such as mobile or embedded devices. To organize our analysis, we focus on three questions: how input-adaptive optimizations (e.g., token pruning and early halting) create attack surfaces for efficiency degradation; how such attacks operate in practice and which optimizations are most vulnerable; and which defenses exist today and whether they meaningfully restore efficiency under attack.

Picture books and comics have long been used to disseminate hateful narratives because they are easily understood even by children, as exemplified by the notorious Nazi propaganda picture book Der Giftpilz. Recently, frontier text-to-image (T2I) systems such as Gemini and GPT-Image have enabled conversational generation with consistent characters and scenes across turns, making hateful visual stories, namely ordered image groups that collectively convey hateful narratives, cheap and scalable to produce. Although prior work has studied hateful content generation by T2I systems, it focuses on individual images, leaving group-level hateful meaning largely unexplored. We aim to address the gap. Concretely, we introduce HatefulStoryPrompts, comprising 330 multi-turn configurations from 55 hateful stories across two languages and three visual styles, and evaluate five frontier models over 4,950 attempts. Every model completes over 80% of the stories, with the strongest reaching 99.0%. We further evaluate existing moderation systems on HatefulVisualStory, a human-labeled dataset of 969 hateful image sets and 990 benign controls, and find that they frequently miss group-level hateful meaning: dedicated safety models achieve at most 34.9% recall, while a strong vision-language model reaches 67.5%. Finally, we propose complementary proactive and post-generation defenses. An interaction-aware monitor achieves 97.3% recall for prompt-only sessions and 92.6% when the user supplies the first image, while post-generation methods jointly analyzing completed image groups reach 80.2%. Our work shows that, as image generation evolves from isolated outputs to coherent visual narratives, safety must evolve accordingly, from per-image moderation to stateful reasoning over interactions and image relationships.

Autonomous AI agent platforms differ substantially in architecture, security, tool integration, execution, autonomy, and deployment, yet the field lacks a common classification scheme for comparing these design choices. We propose ASTELD, an operational six-axis classification framework for autonomous AI agents: Architecture pattern, Security posture, Tool integration model, Execution paradigm, Level of autonomy and human control, and Deployment topology. ASTELD is constructed by synthesizing prior agent taxonomies with observable platform properties and explicit category-assignment rules. We evaluate its discriminative and explanatory utility by mapping eight representative frameworks and by using OpenClaw as an in-depth case study. The resulting profiles separate all eight platforms under their dominant configurations and reveal three cross-platform patterns: a security-accessibility diagonal, strong execution-architecture coupling, and capability convergence with persistent architectural differentiation. We further classify 50+ OpenClaw derivatives and find that innovation concentrates on the Security, Execution, and Deployment axes, indicating that ASTELD can explain where ecosystem fragmentation occurs. The OpenClaw case study also supplies a six-category vulnerability taxonomy, evidence from five institutional assessments, and adoption and governance analyses that connect platform coordinates to observed risks. These results position ASTELD as a reproducible method for comparing agent platforms, identifying unoccupied design regions, guiding framework selection, and organizing future empirical research. The analysis also exposes a consequential empty region: none of the evaluated systems combines local-first deployment with enterprise-grade security.

Prompt injection poses significant security risks to LLM agents. Efficient and effective red-teaming is therefore critical, both for evaluating these risks and for collecting training data to improve defenses. Existing state-of-the-art prompt injection red-teaming methods primarily rely on reinforcement learning (RL), producing attacker models that often generalize poorly to new target LLMs. In this work, we develop PIMiner, an agentic system for prompt injection red-teaming. During training, PIMiner is trained on a sequence of (dataset, target model) pairs and builds a strategy library from scratch. At test time, the learned strategy library can be directly transferred to a previously unseen target LLM without additional training. PIMiner requires only a small number of queries to a target agent (e.g., 10) per test sample. Experimental results demonstrate that PIMiner achieves strong performance. On IPIArena, it attains a 76.2% ASR against Gemini-2.5-Pro, 61.9% ASR against GPT-5.1, and 42.9% ASR against Claude-Sonnet-4.5. On AgentDojo, it achieves an 86.7% ASR against Gemini-2.5-Pro, 53.3% ASR against GPT-5.1, and 40.0% ASR against Claude-Sonnet-4.5.

The semiconductor industry is undergoing a dual revolution: the shift toward heterogeneous 2.5D chiplet systems and the integration of Large Language Models (LLMs) into Electronic Design Automation (EDA) flows. While these paradigms offer unprecedented benefits in yield, modularity, design productivity, etc., they radically expand the hardware attack surface. This paper provides a unified analysis of these frontiers, ranging from attacks on chiplet systems (including hardware stacks for LLM acceleration) across architectural, logical, and physical levels, to various exploits against LLM-driven EDA pipelines. To secure chiplet systems, we review a powerful defense approach that leverages 2.5D split manufacturing and active interposers for physically isolated Root of Trust (RoT) architectures. To secure LLM-driven EDA pipelines, we first identify native threats and then review state-of-the-art defense techniques. Finally, we discuss how LLM systems can advance hardware security efforts for modern systems, including chiplets.

Physical design (PD) intellectual property (IP) is a valuable artifact of modern VLSI implementation. It includes optimized cell placement, clock distribution, and routing decisions produced by carefully tuned PD flows. As access to PD tools expands, unauthorized reuse of placed-and-routed databases becomes an increasing concern. Existing PD watermarking methods either protect only one PD stage or rely on hidden construction details, leaving them vulnerable to a white-box adversary. In this work, we develop PDMarks, a Kerckhoffs-compliant watermarking framework whose security depends only on a secret key. PDMarks embeds ownership evidence across multiple stages of the PD flow, including placement, clock tree synthesis (CTS), and routing. All watermark instances and target values are deterministically derived from a 32-byte secret key using HMAC-SHA256, enabling consistent embedding and verification. PDMarks has been integrated into OpenROAD-flow-scripts. Experiments on NanGate45 and ASAP7 designs show that PDMarks outperforms prior physical design watermarking methods by providing much stronger ownership evidence with comparable or smaller PPA overhead. The approximate joint all-stage coincidence probability is below 10^{-32} for every evaluated design. Wrong-key and attack evaluations further show that incorrect keys do not reproduce the complete ownership proof and that weakening the watermark requires broad perturbation of the protected implementation.

Released aligned large language models remain vulnerable to malicious downstream finetuning. Existing defenses are largely designed for the fine-tuning-as-a-service (FTaaS) paradigm or rely on downstream users to follow additional safety procedures, and therefore do not directly address the setting we study: a provider controlled partially protected open-weight (PPOW) release setting in which most weights remain trainable while a small safety-critical component is preserved at release. We propose a Unidirectional Safety Gate (USG), instantiated as a Null Space Cubic Layer together with an Inverse Adapter inserted after the final Transformer layer. During downstream fine-tuning, the cubic layer suppresses or blocks gradients from harmful samples whose hidden states fall in a calibrated protected region, while the Inverse Adapter restores the base model's forward behavior. In practice, we calibrate a threshold using defender-held harmful data, allowing protection to generalize to nearby in-distribution harmful samples. Across six evaluated model-dataset settings, USG keeps post-finetuning attack success rate close to the pre-release level under a fixed release threshold, while maintaining high safe-pass rates on easier settings and exhibiting a clearer safety-utility trade-off on unsafe samples from BeaverTails. These results suggest that release-time representation-space blocking can raise the cost of malicious downstream adaptation without requiring downstream cooperation. The code is available at https://github.com/OpenCausaLab/Gradient-Immunity.

Direct preference optimization (DPO) is now a standard method for aligning large language models (LLMs) using human preference data. Each DPO example contains a prompt and a pair of candidate model responses. While prompts and responses are often public or model-generated, the relative preference between responses reflects subjective judgments and can reveal sensitive attributes of annotators or end users. Off-the-shelf privacy-preserving approaches are not well matched to this structure, leading to unnecessary noise injection and biased updates in training. In this paper, we formalize preference privacy, a label-DP-style privacy notion for DPO that protects only the relative preference between candidate responses, assuming an adversary who already knows the prompt and responses. We then design PrivDPO, a DPO variant that enforces preference privacy while remaining compatible with large-scale LLM training. Our main observation is that, for neighboring examples differing only in their preference signal, the gradient difference lies on a one-dimensional preference axis determined solely by the text; all preference information flows through this axis. PrivDPO adds calibrated randomness only along this axis via an unbiased randomized rescaling of the DPO objective, avoiding per-example gradient operations. Our experiments on three alignment benchmarks and three LLM families show that PrivDPO consistently achieves strong privacy-utility trade-offs compared with privacy-preserving baselines.

Services increasingly deploy public foundation models with private parameter-efficient adaptations, creating a differential information leakage risk when auditors or adversaries can execute the public base model locally and observe victim outputs. We present VectorHijack-SR, a measurement methodology that converts paired victim/base residuals into calibrated structural bounds over PEFT family, layer locality, and coarse rank, while separating metadata visibility from open-world validity and operational exploitability. Our estimator aggregates query-level magnitude, ranking, entropy, margin, length, template, locality, and spectral statistics into service-level representations. A service-disjoint classifier quantifies structural evidence, and a cross-fitted hierarchical rejector evaluates whether a victim lies outside the calibrated LoRA manifold. Across classification backbones, family leakage exceeds uniform chance on BERT/MNLI (8/12), RoBERTa/MNLI (21/24), and DeBERTa-v3 on MNLI (12/18) and AG News (15/18). Rank inference is task dependent: BERT/MNLI and DeBERTa/AG News reach 8/9, whereas DeBERTa/MNLI achieves 4/9 and is statistically compatible with chance after correction. On a ten-seed BERT open-set grid, the rejector achieves pooled AUROC 0.804 (95% CI [0.660, 0.927]) and known accuracy 0.956, but remains limited on structurally close DoRA and LoRA+head variants. Exact-version linkage on five held-out LoRA-r64 services reaches AUC 0.940. Experiments reveal a visibility-exploitability gap: two-stage recovery provides no fair-budget query savings, posterior-selected PEFT underperforms distill-then-convert PEFT (0.356 vs. 0.517), and free-running generation remains near chance. These results show that known-base, rich-output PEFT services can leak actionable structural and version information, while closed-set confidence alone does not establish universal adapter recovery.

Centralization in maximal extractable value (MEV) markets is a significant concern for blockchain systems, as persistent concentration of economic power can weaken competition, reduce openness, and undermine the decentralization goals of permissionless protocols. While much of the existing analysis has focused on builders, validators, and block-building markets, this paper studies centralization from the perspective of searcher competition. We develop a heterogeneous model in which searchers differ in opportunity coverage and execution efficiency, and we analyze how auction design affects fairness, decentralization, and security among searchers competing for the same MEV opportunity. To evaluate searcher competition, we introduce two metrics: a Shapley-weighted Jain fairness index, which measures whether rewards are proportional to searchers' marginal contributions, and an expected-reward Herfindahl-Hirschman Index (HHI), which measures concentration in long-run searcher rewards. Using these metrics, we first analyze the standard first-price, winner-take-all auction as a benchmark. Our analysis shows that, under searcher heterogeneity, first-price competition can reward rank dominance rather than marginal contribution, leading to concentrated rewards and weaker contribution-adjusted fairness. Motivated by these limitations, we propose an entry-filtered Shapley-capped auction mechanism that distributes searcher rewards more fairly and broadly among admitted high-quality submissions. Designing such a mechanism in a permissionless blockchain environment is challenging: searchers may create Sybil identities by submitting copied or degraded versions of the same execution strategy, and validators may collude with searchers to increase joint payoff. We address these concerns through Bayesian security constraints for copied-code Sybil deviations and validator-searcher coalition deviations.

Digital services increasingly rely on claims that a person, device, or asset was in a specific place at a specific time. Today, those claims often depend on self-reported location data, which is easy to falsify and difficult to verify after the fact. Proof-of-Location (PoL) systems address this gap by turning presence claims into evidence that an independent verifier can later inspect. This paper builds upon recent theoretical work on decentralized PoL architectures and demonstrates how they can move from emulation to a physical prototype built with low-cost hardware. We implement a witnessing zone in which fixed nearby devices measure a prover's presence, exchange claims over a local mesh, and record them in a tamper-evident ledger. Building the prototype required adapting the abstract protocol to physical constraints through witness-initiated ranging, cross-witness consistency checks, and freshness binding against replay. Our controlled indoor evaluation shows that the system can produce accurate, low-latency proof objects while detecting simulated replay and malicious-ranging attacks. The result is a reusable experimental baseline for next-generation digital trust infrastructure, that exposes the remaining calibration, verifier-independence, radio-integrity, and scaling requirements for decentralized location evidence, where physical presence claims can be independently checked under real radio, networking, and timing conditions.

Hardware designs, like software, are susceptible to bugs that can introduce security vulnerabilities and create opportunities for malicious exploitation. Unlike software vulnerabilities, however, hardware flaws become permanently embedded in silicon after fabrication, making them difficult or impossible to patch. Many of these weaknesses are categorized under the Common Weakness Enumeration (CWE) framework and include improper access control, exposure of sensitive information, and unintended privilege escalation. To improve the detection of such vulnerabilities, we propose a methodology that leverages a Large Language Model (LLM) to identify potential hardware CWEs directly from hardware designs in Verilog. The proposed approach is evaluated iteratively on a dataset of single-module Verilog designs to assess its effectiveness in detecting hardware security weaknesses. Our results demonstrate the potential of LLMs to augment traditional hardware security analysis by providing automated, scalable assistance for identifying security vulnerabilities during the hardware design process.

Multi-agent LLM systems relay key-value caches instead of text and credit their gains to exchanged "latent thoughts". That credit is a claim about which example's cache is relayed, not merely that one is. We audit it causally in released systems. The cache is replaced with deranged (mismatched-example), zeroed, and moment-matched random counterparts, under two regimes defined by whether the receiver needs the sender's private information. Where it does, the battery reads ceiling: 100% against 23-25% for answer-irrelevant relays on the primary backbone, a contrast replicated across three families, five checkpoints, and a prose document-QA surface. Where it does not, a pre-registered five-seed protocol establishes equivalence within 2.8 points, a margin anchored to the audited system's reported gain, under Holm-corrected TOST on GSM8K and ARC-Challenge across three Qwen3 scales and on MedQA at 8B (one cell shows a small detected advantage inside the margin); a second family shows no detected advantage. A large cache effect need not be a pairing effect. In one natural cell, zeroing the relay costs 14.7 points; a mismatched cache, 0.4. Nor is need sufficient: under the same test, delivered channels span ceiling (LatentMAS's native relay), partial (KVComm's layer subset), and no detected example-specific transfer (C2C's released projector). Benchmark deltas do not by themselves establish latent-thought transmission; establishing it takes a mismatched-cache audit, which we release.

Radio frequency fingerprint identification (RFFI) is a promising technique for wireless device authentication. However, practical RFFI systems must enroll newly authorized devices while rejecting previously unseen ones, even when the feature extractor is trained on only a few labeled base-device classes, giving rise to a high-openness RFFI problem. Existing open-set recognition methods typically rely on feature spaces learned from a large and diverse set of known-device classes, limiting their applicability in practical scenarios. To address this challenge, we propose HoRFFI, a high-openness RFFI framework that supports scalable device identification and unknown-device rejection using only a small number of labeled training devices. HoRFFI employs a similarity-enhanced variational information bottleneck (SVIB)-based supervision mechanism, which reduces the encoder's dependence on training-class diversity and learns a more transferable embedding space. This supervision mechanism uses feature-space augmentation and clustering to derive inter-sample similarity information, which provides supplementary supervision for regularizing the embedding space. Experiments on public LoRa and Wi-Fi datasets show that HoRFFI achieves absolute improvements of \(0.112\) and \(0.288\) in novel-class accuracy, respectively, and corresponding absolute AUC improvements of \(0.029\) and \(0.060\) over the best-performing baselines.