Papers

Security operations centers (SOCs) are beginning to use large language models (LLMs) as copilots to draft incident-response plans. These plans may include actions that are valid per the catalog but still violate mandatory steps, required ordering, or approval gates before analyst review. SOCpilot makes this compliance question measurable at the plan boundary. It fixes the incident package, action catalog, policy rules, verifier, and public evidence surface. Next, it verifies the copilot's proposed action trace. We evaluate two LLM providers on 200 real incidents from an anonymized production SOC in a financial-sector case study. We compare their plans to paired analyst-authored references from the same security orchestration, automation, and response (SOAR) cases. An identical inline policy text moves the two providers in opposite directions. A deterministic verifier removes 466 non-compliant, approval-gated actions, without reducing baseline-task recall. Aggregate rates remain stable across 3 reruns of the fixed corpus. The official evidence focuses on approval-gated decisions regarding recovery and containment. Separately, the artifact exposes zero-cost readiness checks for mandatory and ordering repairs. We release the runnable artifact so independent reviewers can rederive the public results without access to private incident data.

This work introduces PAS – Privacy Anchor Substitution, a structured mechanism for enabling user location privacy in spatial retrieval-augmented generation (RAG) systems. Unlike conventional differential privacy methods that directly perturb user locations, PAS represents location with relative anchor encoding consisting of an anchor, direction bin, and distance bin, allowing seamless integration with modern RAG pipelines. We evaluate PAS on a synthetic urban dataset and show that it achieves impressive coarse privacy guarantees, with approximately 370-400m adversarial location error, while retaining more than half of the baseline retrieval performance. Despite the slight drop in retrieval performance, the downstream generation quality under PAS remains comparatively robust, indicating that large language models can compensate for imperfect spatial retrieval. Furthermore, we provide empirical analysis showing that PAS exhibits non-monotonic privacy-utility relationship with respect to privacy parameters. We attribute this to geometric bias induced by anchor discretization, making it different from continuous noise mechanisms such as geo-indistinguishability. Our results show that structured spatial representations offer a practical approach to privacy in location based reasoning in RAG systems.

Proper implementation of technical and administrative controls reinforces an organization's cybersecurity posture and business resilience, reduces risks, and enhances governance, ultimately elevating business maturity. The dynamics of the technological landscape and emerging threats negatively affect the most diverse companies, regardless of their size. This, associated with a global gap in the cybersecurity workforce, imposes enormous challenges and the need for a profound change in how companies respond to threats. Generative Artificial Intelligence from large language models has become an influential tool across various companies, emerging as a viable option to help address those challenges while partially addressing the shortage of skilled labor. Although large language models can help in this scenario, there may be risks, such as generating unreliable or 'hallucinated' content, which could lead people and companies to make bad decisions. Our study proposes integrating human experts into the validation process as a crucial step toward ensuring the proper implementation of technical and administrative controls. Furthermore, we sought to identify how large language models perform in assessing cybersecurity risk scenarios compared to human experts, highlighting the importance of integrating humans and machines in the cybersecurity risk assessment process. Using a questionnaire with risk scenarios, we analyzed responses from 50 human experts. We compared their responses with those of five popular large language models to determine whether it is possible to use only large language models for cybersecurity risk assessment. The results reveal that the large language models consistently underestimated cybersecurity risks compared to human experts, reinforcing the need for human oversight and suggesting that LLMs should be used as complementary tools rather than standalone assessors.

Large language models (LLMs) remain vulnerable to adversarial prompting despite advances in alignment and safety, often exhibiting harmful behaviors under novel attack strategies. While adversarial training can improve robustness, existing approaches are computationally expensive and difficult to scale. Recent continuous adversarial training methods, such as Continuous adversarial training (CAT) and Continuous Adversarial Preference Optimization (CAPO), address this challenge by leveraging gradient-based perturbations in the embedding space, enabling more efficient and expressive attacks. Building on this paradigm, we propose WARDEN, a distributionally robust adversarial training framework for LLMs that dynamically reweights adversarial examples through an f -divergence ambiguity set around the empirical training distribution. Our method optimizes the worst-case adversarial loss within a divergence ball around the empirical data distribution, automatically emphasizing harder adversarial examples. Using the convex dual formulation, the objective reduces to a log-sum-exp form under the KL divergence, with a dynamical parameter controlling the strength of reweighting. This study leads to a new class of information-theoretic objectives that significantly reduce attack success rates while maintaining model utility. Across multiple LLMs and attack settings, WARDEN substantially reduces attack success rates with computational and utility costs comparable to CAT-, CAPO-, and MixAT-based baselines, making it a practical approach for scalable robust alignment.

Log-based detection rules remain central to modern security operations, encoding domain expertise that analysts iteratively refine to balance detection coverage against alert volume. Yet while prior work has examined the evolution of network intrusion detection signatures, the longitudinal behavior of log-based detection rules has received little empirical study. We present the first longitudinal analysis of detection rule evolution across two widely used repositories: the community-driven Sigma project and the curated Splunk Security Content (SSC). To compare rule versions based on detection logic rather than surface syntax, we introduce a predicate graph intermediate representation that canonicalizes the logical structure of a rule, together with a tree alignment procedure for analyzing changes across revisions. We apply this method to 6,859 rule histories from Sigma and SSC and find that roughly 56% of rules undergo at least one revision on detection logic. Across rule lifetimes, evolution is predominantly non-monotonic, with over half of rules both adding and removing clauses over time. We further observe recurring reversions, indicating that changes are often revisited rather than strictly accumulated. Combining structural analysis with LLM-based inference and human validation of operational intent shows that roughly a quarter to a third of rules alternate between expanding coverage and reducing false positives, rather than converging toward a stable form. Together, these results reveal that detection rule evolution in public repositories reflects ongoing operational trade-offs rather than steady convergence. Our study raises questions about why rules change the way they do and supports research towards better processes for devising and deploying security rules.

As Vision-Language Models (VLMs) are increasingly deployed as autonomous cognitive cores for embodied assistants, evaluating their privacy awareness in physical environments becomes critical. Unlike digital chatbots, these agents operate in intimate spaces, such as homes and hospitals, where they possess the physical agency to observe and manipulate privacy-sensitive information and artifacts. However, current benchmarks remain limited to unimodal, text-based representations that cannot capture the demands of real-world settings. To bridge this gap, we present ImmersedPrivacy, an interactive audio-visual evaluation framework that simulates realistic physical environments using a Unity-based simulator. ImmersedPrivacy evaluates physically grounded privacy awareness across three progressive tiers that test a model's ability to identify sensitive items in cluttered scenes, adapt to shifting social contexts, and resolve conflicts between explicit commands and inferred privacy constraints. Our evaluation of 12 state-of-the-art models reveals consistent deficits. In cluttered scenes, all models exhibit monotonic performance decay as scene complexity grows due to perceptual deficit. When social context shifts, no model exceed 65% selection accuracy. Under conflicting commands, the best model gemini-3.1-pro perfectly balances task completion and privacy preservation in only 51% of cases. These findings reveal that current VLMs in the physical world suffer from perceptual fragility and fail to let their knowledge of privacy cues govern their situated behavior. Our code and data is available at https://github.com/immersed-privacy/immersed-privacy .

Retrieval-Augmented Generation (RAG) and agentic AI systems are increasingly prevalent in enterprise AI deployments. However, real enterprise environments introduce challenges largely absent from academic treatments and consumer-facing APIs: multiple tenants with heterogeneous data, strict access-control requirements, regulatory compliance, and cost pressures that demand shared infrastructure. A fundamental problem underlies existing RAG architectures in these settings: retrieval systems rank documents by relevance–whether through semantic similarity, keyword matching, or hybrid approaches–not by authorization, so a query from one tenant can surface another tenant's confidential data simply because it scores highest. We formalize this gap and analyze additional shortcomings–including tool-mediated disclosure, context accumulation across turns, and client-side orchestration bypass–that arise when agentic systems conflate relevance with authorization. To address these challenges, we introduce a layered isolation architecture combining policy-aware ingestion, retrieval-time gating, and shared inference, enforced through server-side agentic orchestration. This approach centralizes security-critical operations–tool execution authorization, state isolation, and policy enforcement–on the server, creating natural enforcement points for multitenant isolation while allowing client-side frameworks to retain control over agent composition and latency-sensitive operations. We validate the proposed architecture through an open-source implementation in OGX, a vendor-neutral framework that implements an OpenAI-compatible, open-source Responses API with server-side multi-turn orchestration. We evaluate it empirically and show that ABAC gating eliminates cross-tenant leakage while introducing negligible overhead.

Production LLM systems require both safety moderation and PII detection under strict latency and cost constraints. This creates a trade-off: autoregressive moderators are accurate but expensive, while lightweight encoders are faster but less capable. We present GLiNER Guard (GLiGuard), a unified encoder that performs safety classification and PII detection in a single forward pass, simplifying safety pipelines. We introduce three variants: compact uni- and bi-encoders (145-147M) for high-throughput serving, and GLiGuard Omni (209M) for stronger moderation quality. Under dynamic batching on a single A100, the compact model reaches 193 requests/sec with P99 latency below 1s, achieving 1.6x higher throughput than GLiNER2. Omni remains competitive with much larger moderators on public safety benchmarks. We also release PII-Bench, a span-level benchmark for evaluating PII detection in end-to-end pipelines. Overall, encoder-based guardrails offer a practical low-cost alternative for always-on moderation. Models and benchmarks are released on HuggingFace.

Network-based Intrusion Detection Systems (IDS) are predominantly trained on tabular flow records, whose one-dimensional representations limit convolutional architectures from exploiting inter-feature spatial correlations. This paper presents a novel byte-level flow-to-image encoding method that converts each network-flow record into a fixed-size RGB image. Continuous features are serialised using IEEE-754 single-precision format and packed sequentially into pixels along an inverted-L shaped trajectory, while discrete features are mapped to byte values and placed contiguously in the middle image row's centre. The encoding is deterministic and reversible, preserving a fixed spatial layout across all samples. Four IDS models are evaluated on NSL-KDD and UNSW-NB15 datasets with both flow and image-based configurations. The image-based representation yields consistent accuracy gains of up to 15.6% and 12.8% for binary and multi-classification on UNSW-NB15, and up to 3.5% and 3.2% on NSL-KDD, highlighting the potential of byte-level visual encoding to strengthen AI-driven intrusion detection in local computer networks.

Large language model (LLM) ecosystems such as Claude Code and ChatGPT increasingly rely on skills: packages of natural-language instructions and executable tools. Once in the LLM's context, skill content cannot be reliably separated from trusted instructions, and a skill's executable side can invoke privileged actions, exposing the skill supply chain to injection, tampering, and rug-pull attacks. Existing defenses are stage-bound: centralized signing, audit reports unbound from the runtime artifact, or policy engines that cannot attest to what was approved. We present SIGIL, the first framework that seals the audit-runtime gap for LLM skills. SIGIL delivers verifiable hosting through a tamper-evident, decentralized on-chain registry from which LLMs fetch skills directly. The registry admits four publication types, Transparent, Licensed, Sealed, and Committed, spanning plaintext public distribution, monetized access, custodial use, and off-chain workflows; before admission, every skill is vetted by a Decentralized Autonomous Organization (DAO) audit committee that supports pluggable auditing methods under a stake-and-slash economic model. At load time, SIGIL delivers verified loading through a skill verification protocol executed by a Skill Verification Loader (SVL) embedded as the mandatory loading path: the SVL retrieves and decrypts the skill as its type requires, verifies its integrity against the on-chain record, and enforces its permission manifest before context injection. We evaluate SIGIL on a real-world deployment against 1,023 in-the-wild skills spanning six attack types. At load time, the SVL verifies each skill's integrity against its on-chain record and enforces its approved permission manifest, completing batched verification under 86 ms. Together, these results show that LLM skills can be cryptographically bound from publication through runtime at practical cost.

LLMs become increasingly capable, editorial boards and program committees are growing concerned about reviewers who fully outsource peer review to commercial chatbots. This concern stems from prior findings that current chatbots lack the independent critical thinking and depth of reasoning required to assess scientific novelty. One promising direction for mitigating this concern is to embed hidden instructions into manuscripts that disrupt or alter chatbot-generated reviews. However, existing methods remain intuitive and fragile, as they typically rely on homogeneous payloads injected in an inter-stream manner, rendering them susceptible to sanitization or neutralization. More broadly, the community still lacks a systematic formulation of this threat and a principled defense framework. In this paper, we identify End-to-End Review Outsourcing as an emerging threat and propose IntraGuard, a black-box, venue-agnostic defense framework grounded in the structural–visual decoupling inherent to the PDF. Designed for committee-side deployment, IntraGuard supports both explicit strategies that trigger refusal or warning signals, and implicit strategies that embed predefined textual markers into the generated review. These strategies can be deployed via any of three intra-stream injection mechanisms, each of which seamlessly embeds heterogeneous defensive text objects within the PDF's underlying structure without altering its visual presentation. Extensive evaluations (over 17,844 cases) across 7 real-world commercial chatbot settings and 12 venues spanning diverse disciplines show that IntraGuard achieves a defense success rate of up to 84%, while preserving peer-review invariance for human reviewers. We further evaluate 11 adaptive attacks spanning manuscript sanitization and instruction interference, and discuss the implications of constructing ensemble defenses.

As 6G architectures transition toward decentralized Multi-Agent Systems (MAS), ensuring secure access to shared Knowledge Bases (KB) is critical. Traditional authorization models like RBAC fail to provide the granularity required for autonomous agents interacting with Semantic-based data. This work proposes a hybrid authorization framework that integrates roles and First-Order Logic (FOL) predicates to enforce zero-trust principles at the knowledge-graph level. We eliminate permission inheritance by enforcing authorization at the triple level (Subject-Predicate-Object), ensuring agents only access metadata required for their specific functional lifecycle.

We consider the extensive-form bandit problem, where on each trial the learner (a user coordinated by a server) plays an extensive-form game against an oblivious adversary, observing the information sets it finds itself in as well as the resulting payoff/loss. We give an algorithm for this problem that satisfies $\epsilon$-local differential privacy and attains a regret of $\tilde{O}(\sqrt{A\ln(S)T}/\epsilon)$, where $A$ is the total number of actions that the learner can possibly take, $S$ is the number of the learner's possible reduced strategies, and $T$ is the number of trials. On each trial, the time complexity of our algorithm is, up to a factor logarithmic in the maximum number of actions at an infoset, equal to the time required for the server to transmit the reduced strategy to the user. We note that local differential privacy is the strongest version of differential privacy and, to the best of our knowledge, this is the first work to study differential privacy of any form in the extensive-form bandit problem.

The deployment of Large Language Model (LLM)-generated SQL queries in Artificial Intelligence of Things (AIoT) systems introduces critical security risks, as prompt injection attacks can manipulate LLMs into producing unauthorized queries that expose sensitive data or execute destructive operations. Existing NL2SQL research focuses on query accuracy, while MCP server implementations provide only SQL-level protections without fine-grained role-based access control. This paper proposes SecureMCP, a policy-enforced LLM data access framework integrating Role-Based Access Control (RBAC) with an MCP server to establish multi-layer defense for LLM-generated SQL execution. The framework incorporates five defense modules – check_policy for table-and-column-level RBAC, explain_gate for cost-explosive query blocking, SQL Interceptor for dangerous pattern detection, Risk Level Filter for SQL risk classification, and DB Isolation for cross-database restriction – operating in a sequential fail-closed pipeline mapped to six prompt injection types grounded in the OWASP Top 10 for LLM Applications. We evaluate SecureMCP on the IoT-SQL dataset (11 tables, 173 columns, 239,398 records) using Qwen3-8B. Experiment A demonstrates that defense modules preserve execution accuracy, with EX-in-ALLOW remaining within 65.1%-76.4% across four RBAC roles, matching the unprotected baseline of 63.8%. Experiment B shows that SecureMCP achieves 82.3% Policy Compliance on 2,400 adversarial queries, with genuine defense failures limited to 3.4%. The defense-in-depth analysis reveals check_policy accounts for 78.7% of blocks, while secondary modules contribute an additional 17.5 percentage-point improvement. The Injection Incorporation Rate of 72.5% confirms high LLM susceptibility, establishing the necessity of external policy enforcement.

Decentralized finance (DeFi) protocols now intermediate over USD 100 billion in value, including regulated stablecoins and tokenized assets deployed as collateral, yet no widely adopted framework operationalizes risk assessment at the rigor institutional adoption demands. Existing approaches emphasize protocol-specific parameter optimization or conceptual taxonomies without providing explainable, composability-aware, and structurally independent assessment methodologies. We propose a nine-dimension DeFi risk assessment framework extending the six-dimension taxonomy introduced by Moody's Analytics and Gauntlet with three novel dimensions: composability risk, comprehension debt, and temporal risk dynamics. We additionally introduce a transparency confidence modifier separating assessment reliability from risk severity. The framework is grounded in structural analysis of protocol dependencies conducted through an ontology-based protocol intelligence infrastructure covering more than 8,000 DeFi protocols. We retrospectively analyze 12 major DeFi-related incidents from 2024-2026 representing approximately USD 2.5 billion in direct losses. Five of the 12 incidents require at least one novel dimension for complete root-cause characterization, including the two highest-systemic-impact events in the dataset.

Large Language Models (LLMs) have achieved remarkable success but remain highly susceptible to jailbreak attacks, in which adversarial prompts coerce models into generating harmful, unethical, or policy-violating outputs. Such attacks pose real-world risks, eroding safety, trust, and regulatory compliance in high-stakes applications. Although a variety of attack and defense methods have been proposed, existing evaluation practices are inadequate, often relying on narrow metrics like attack success rate that fail to capture the multidimensional nature of LLM security. In this paper, we present a systematic taxonomy of jailbreak attacks and defenses and introduce Security Cube, a unified, multi-dimensional framework for comprehensive evaluation of these techniques. We provide detailed comparison tables of existing attacks and defenses, highlighting key insights and open challenges across the literature. Leveraging Security Cube, we conduct benchmark studies on 13 representative attacks and 5 defenses, establishing a clear view of the current landscape encompassing jailbreak attacks, defenses, automated judges, and LLM vulnerabilities. Based on these evaluations, we distill critical findings, identify unresolved problems, and outline promising research directions for enhancing LLM robustness against jailbreak attacks. Our analysis aims to pave the way towards more robust, interpretable, and trustworthy LLM systems. Our code is available at Code.

LLM agents have been increasingly adopted for solving security tasks. However, existing evaluations usually require source code access, while commercial off-the-shelf (COTS) binaries dominate deployed software and require reasoning from stripped, optimized machine code. This discrepancy raises an important question: can modern LLM agents reason about vulnerabilities in critical COTS binaries? Motivated by this question, we build SLYP, a REACT-style pipeline for end-to-end vulnerability discovery and validation of COTS binaries. SLYP combines extensible MCP servers for binary exploration and dynamic debugging, and validates candidate vulnerabilities by synthesizing debugger-verified proof-of-concept (PoC) crashes. We evaluate SLYP and production coding agents, including Claude Code and Codex, on COTS Windows binaries centered on a 20-object COM benchmark. SLYP uncovers all 64 vulnerable entry functions while default production agents miss up to 15; SLYP also surfaces more true vulnerabilities than the state-of-the-art static analyzer, which discovers at most 35 with a large number of false positives. For validation, SLYP generates debugger-verified PoCs for 67.5% of cases, while default production agents generate none. Further ablations show that tool sets and model choice materially affect COTS binary reasoning. Our additional evaluation also demonstrates the generalizability of SLYP on Windows kernel targets. To date, SLYP has uncovered 39 zero-day vulnerabilities, 31 in COM/RPC services and 8 in kernel drivers, all disclosed to the Microsoft Security Response Center (MSRC), with 23 assigned CVEs and $203,000 bounty awards.

Public repositories now distribute thousands of specialized Low-Rank Adaptation (LoRA) modules, but a third-party adapter routinely arrives without the refusal behavior the same adapter would have had if it had been trained responsibly. Restoring it by fine-tuning on safety data induces the opposite failure: the domain expertise the adapter was published to provide degrades. A practitioner who downloads a finished adapter holds neither the domain corpus nor a safety corpus, so neither remedy is available. We propose Neural Weight Translation (NeWTral), which maps unsafe domain adapters to their safety-aligned counterparts entirely within parameter space while preserving their expertise. NeWTral is a non-linear translator pre-trained on unsafe-to-safe adapter pairs, with a Mixture of Experts router that blends a high-fidelity surgical translator and an aggressive alignment expert per tensor. Across four architectural families (Llama, Mistral, Qwen, Gemma) at scales up to 72B parameters and eight professional domains, NeWTral cuts the average Attack Success Rate from 70% to 13% while retaining 88% knowledge fidelity. The effect reproduces on JailbreakBench, on 12 of 14 adapters downloaded from a public model hub, and under automated jailbreak attacks. Curing is a single pass over the weights, needing no original training data and no retraining.

Atomic swaps are a fundamental primitive for the trustless exchange of digital assets across blockchains: they guarantee that either both parties receive the agreed assets or neither party transfers. While this all-or-nothing guarantee is powerful, it also imposes an inherent determinism that rules out exchanges whose intended outcome is probabilistic. As a result, existing atomic swaps cannot realize trustless exchanges in which one party pays for a fixed chance of receiving a larger asset or reward, as in lotteries, randomized allocation mechanisms, and probabilistic cross-chain trades. We introduce probabilistic swaps, a new cryptographic primitive that extends atomic swaps to the probabilistic setting. In a probabilistic swap, one party's transfer is executed with a fixed, publicly specified probability embedded in the protocol and cannot be biased by either party. This yields a trustless mechanism for randomized exchange with verifiable odds and no trusted intermediary. Our construction combines adaptor signatures with oblivious pseudorandom functions (OPRFs) to realize the desired probabilistic outcome while ensuring that neither party can predict or bias it in advance. Along the way, we introduce a new mechanism for the atomic exchange of OPRF evaluations for payments, which may be of independent interest. A key feature of our approach is that it preserves the minimal on-chain footprint of modern atomic-swap protocols. The protocol relies only on standard Bitcoin scripts, such as digital signatures and timelocks, and is deployable on any blockchain that already supports atomic swaps. Consequently, probabilistic swaps are indistinguishable from ordinary on-chain transactions, which helps preserve privacy and fungibility. We provide formal security foundations and demonstrate practicality through a probabilistic swap in the Bitcoin testnet and in the Lightning Network.

For Transformer models, cryptographically secure inference ensures that the client learns only the final output, while the server learns nothing about the client's input. However, securely computing nonlinear layers remains a major efficiency bottleneck due to the substantial communication rounds and data transmission required. To address this issue, prior works reveal intermediate activations to the client, allowing nonlinear operations to be computed in plaintext. Although this approach significantly improves efficiency, exposing activations enables adversaries to extract model weights. To mitigate this risk, existing works employ a shuffling defense that reveals only randomly permuted activations to the client. In this work, we show that the shuffling defense is not as robust as previously claimed. We propose an attack that aligns differently shuffled activations to a common permutation and subsequently exploits them to extract model weights. Experiments on Pythia-70m and GPT-2 demonstrate that the proposed attack can align shuffled activations with mean squared errors ranging from $10^{-9}$ to $10^{-6}$. With a query cost of approximately \$1, the adversary can recover model weights with L1-norm differences ranging from $10^{-4}$ to $10^{-2}$ compared to the oracle weights.

As security demands increase, the importance of secure computation technologies grows, yet these technologies can often seem overwhelming to practitioners. Furthermore, many approaches focus only on a single technology, potentially overlooking superior alternatives. This work aims to address the issue of selecting the right technology for secure computation by presenting a comparative analysis of two highly relevant cryptographic methods and their software implementations, with a particular focus on machine learning. Firstly, we provide a theoretical summary and comparison of the secure computation paradigms of secure multi-party computation (SMPC) and fully homomorphic encryption (FHE). We outline the advantages and limitations of the protocols, as well as the relevant open-source software implementations. Secondly, we present the results of extensive benchmarking of the main software frameworks identified for machine learning operations and models. Regarding the current state of the art in FHE, we observe that it outperforms SMPC for regressions. Additionally it may be faster for simple dense networks using GPUs or Hybrid Models. Conversely, SMPC showed superior performance for complex models such as CNNs. Our results should pave the way for more technology-agnostic benchmarking of secure computation technologies for machine learning, providing guidance for practitioners looking to adopt these technologies.

Protecting confidential data while preserving utility is particularly challenging when data sets contain outlying observations. Existing latent space anonymization methods, such as spectral anonymization (SA), rely on principal component analysis (PCA) and may therefore be vulnerable to contamination. We investigate anonymization in the presence of outliers and propose ICSA, a robust alternative to SA based on invariant coordinate selection (ICS). By replacing the PCA transformation with ICS, the robustness of the anonymization procedure can be regulated through the choice of scatter matrices. Alongside the methodological development, we derive a theoretical result showing that SA fails under sufficiently influential outliers. To assess the practical implications of this result, we compare the privacy-utility trade-off of ICSA and SA through simulation experiments under varying contamination settings and outlier severities. Our findings indicate that implementations of ICSA based on robust scatter matrices achieve stronger privacy protection than SA, while typically maintaining comparable, and in some cases improved, utility. We further examine the empirical performance of the proposed method using a benchmark clinical data set, where ICSA demonstrates superior overall privacy-utility efficiency relative to SA. These results suggest that explicitly accounting for outliers can materially improve anonymization performance and that robust latent space transformations offer a promising direction for privacy-preserving statistical data release.

Replacing conventional devices with smart ones has many advantages, e.g., a seamless integration of physical objects into the users digital environment or improved modes of use. However, if a conventional device is replaced by a smart device, its IT components can cause risks, that shorten the life of the device. Such risks stem from different life cycles of embedded soft- and hardware, libraries and protocols used, and the IT ecosystem required. This is problematic, because many conventional household appliances, say, a fridge or TV, have a much longer life span than typical IT equipment. In this paper, we use a systematic approach to identify long-term risks for the operational life span of a smart fridge. In particular, we identify 8 different use cases of three typical smart fridges, e.g., cooling or managing "best before" dates. We model the IT ecosystem needed to run these use cases, and we inspect each asset in this ecosystem for potential long-term risks. We found that even cooling, the most basic use case, is at risk in the long run. This is because the setting cooling parameters may depend on parts of the IT ecosystem that are not under the users control. On the other hand, we did not find any risk that may lead to harm of the category "threatening". Our findings on the smart fridge can be generalized to other smart devices easily.

Modern AI agents execute real-world side effects through tool calls such as file operations, shell commands, HTTP requests, and database queries. A single unsafe action, including accidental deletion, credential exposure, or data exfiltration, can cause irreversible harm. Existing defenses are incomplete: post-hoc benchmarks measure behavior after execution, static guardrails miss obfuscation and multi-step context, and infrastructure sandboxes constrain where code runs without understanding what an action means. We present AgentTrust, a runtime safety layer that intercepts agent tool calls before execution and returns a structured verdict: allow, warn, block, or review. AgentTrust combines a shell deobfuscation normalizer, SafeFix suggestions for safer alternatives, RiskChain detection for multi-step attack chains, and a cache-aware LLM-as-Judge for ambiguous inputs. We release a 300-scenario benchmark across six risk categories and an additional 630 independently constructed real-world adversarial scenarios. On the internal benchmark, the production-only ruleset achieves 95.0% verdict accuracy and 73.7% risk-level accuracy at low-millisecond end-to-end latency. On the 630-scenario benchmark, evaluated under a patched ruleset and not claimed as zero-shot, AgentTrust achieves 96.7% verdict accuracy, including about 93% on shell-obfuscated payloads. AgentTrust is released under the AGPL-3.0 license and provides a Model Context Protocol server for MCP-compatible agents.

Industrial Control Protocols (ICPs) are critical to the reliability and stability of industrial infrastructure, yet their security is fundamentally compromised by a specification-blindness bottleneck. Modern fuzzers, constrained by observation-driven inference, struggle to penetrate deep protocol states or detect subtle semantic deviations. In this paper, we present AFL-ICP, an autonomous fuzzing framework that pioneers a specification-driven paradigm. AFL-ICP features a context-aware specification formalization pipeline to transform complex specifications into rigorous machine-executable grammars. Building on this formalized specification, AFL-ICP leverages LLMs to enable automated protocol adaptation and seed generation, allowing for rapid extension to new protocols with minimal manual effort. Additionally, it includes an LLM-powered differential checker that cross-references implementation outputs with specification requirements to detect subtle semantic and logic bugs that existing fuzzers cannot detect. We implement AFL-ICP and evaluate it on four widely used ICPs, including both open-source and closed-source variants. Results show that AFL-ICP significantly outperforms state-of-the-art fuzzers in coverage and uncovers 24 previously unknown vulnerabilities, for which we have received acknowledgments from affected vendors (e.g., FreyrSCADA). Specifically, the identified vulnerabilities include 16 semantic and logic bugs that can silently disrupt industrial operations and degrade service availability.

The pervasive integration of AI has enabled Offensive AI: the exploitation of AI for malicious ends across the cyber-kill chain. A critical manifestation is the user attribute inference attack, where AI infers sensitive Personally Identifiable Information (PII) from innocuous public data. We explore how music streaming ecosystems, where users routinely release public playlists, can be exploited for Offensive AI. To quantify this threat, we developed musicPIIrate. This novel tool leverages deep learning architectures that utilize both standalone data representations and the structural information embedded in a user's playlist collection. Our design explores set-based approaches (e.g., Deep Sets) and methodologies modeling relationships between playlists (e.g., Graph Neural Networks), which we also combine to leverage both perspectives. Our approach addresses feature extraction from unordered, variable-length set data, enabling accurate PII prediction. Empirical evaluation demonstrates that musicPIIrate achieves state-of-the-art inference accuracy. The tool successfully infers a wide array of attributes, including: Demographics (Age, Country, Gender), Habits (Alcohol, Smoke, Sport), and Personality Traits (OCEAN scores). musicPIIrate outperforms existing methods, beating baselines in 9 out of 15 attribute inference tasks. To counter this vulnerability, we propose JamShield, a lightweight defensive framework. JamShield strategically injects dummy playlists into an account to dilute the PII-carrying signal. Our analysis indicates that JamShield represents a promising defense, lowering inference F1-scores by an average of 10%. This work provides an initial Offensive-AI benchmark for playlist-based PII inference using architectures that leverage set- and graph-structured data and introduces a defense showing encouraging mitigation effects.

Today, advances in medical technology extensively utilize 3D volume data for accurate and efficient diagnostics. However, sharing these data across networks in telemedicine poses significant security risks of data tampering and unauthorized copying. To address these challenges, this paper proposes a novel reversible-zero watermarking approach, termed Vol-Mark, for medical volume data to protect their ownership and authenticity in telemedicine. The proposed Vol-Mark method offers two key benefits: 1) it designs a volume data feature extractor that leverages contrastive learning to efficiently extract discriminative and stable volumetric features, ensuring robustness against 3D attacks; 2) it introduces the cubic difference expansion (c-DE) technique, which leverages the 3D integer wavelet transform to embed watermark bits into neighboring voxels within cubes at low-frequency coefficients. The voxel differences within each cube are expanded to create embedding space, and a majority voting mechanism is employed during extraction to enhance reliability. The embedding process incurs low distortion and supports lossless removal, thereby preserving the integrity and diagnostic accuracy of medical volume data. Through these two benefits, Vol-Mark enables both integrity verification and ownership verification. Integrity verification is first performed, and ownership verification through hypothesis testing is further conducted to enhance reliability, particularly under data tampering or watermark removal attacks. Comprehensive experimental results show the effectiveness of the proposed method and its superior robustness against conventional, geometric, and hybrid attacks on medical volume data. In particular, through multiple tasks evaluations, Vol-Mark consistently achieves an ACC above 0.90 in most attack scenarios, outperforming existing methods by a clear margin.

Jailbreak attacks on audio language models (ALMs) optimize audio perturbations to elicit unsafe generations, and they typically update the entire waveform densely throughout optimization. In this work, we investigate the necessity of such dense optimization by analyzing the structure of token-aligned gradients in ALMs. We find that gradient energy is highly non-uniform across audio tokens, indicating that only a small subset of token-aligned audio regions dominates the optimization signal. Motivated by this observation, we propose Token-Aware Gradient Optimization (TAGO), which enables sparse jailbreak optimization by retaining only waveform gradients aligned with audio tokens that have high gradient energy, while masking the remaining gradients at each iteration. Across three ALMs, TAGO outperforms baselines, and substantial sparsification preserves strong attack success rates (e.g. on Qwen3-Omni, $\mathrm{ASR}_{l}$ remains at 86% with a token retention ratio of 0.25, compared to 87% with full token retention). These results demonstrate that dense waveform updates are largely redundant, and we advocate that future audio jailbreak and safety alignment research should further leverage this heterogeneous token-level gradient structure.

Modern malware detection pipelines rely on continuous data ingestion and machine learning to counter the high volume of novel threats. This work investigates a realistic gray-box poisoning threat model targeting these pipelines. Using the secml_malware framework, we generate problem-space adversarial binaries through functionality-preserving manipulations, specifically Import Address Table (IAT) and section injections. We evaluate the impact of these poisoned samples when ingested into a defender's training set for a LightGBM malware detection model. Our empirical results demonstrate that subtle IAT-based perturbations enable compact poisoning samples that significantly degrade detection recall. These findings illustrate the inherent challenge of developing low-visibility adversarial perturbations that maintain high poisoning efficacy within continuous learning systems. We further evaluate a defense mechanism based on a homogeneous ensemble, which successfully identifies and filters up to 95.6% of poisoning attempts while maintaining a high retention rate for legitimate data. These findings emphasize the necessity of robust pre-ingestion validation in production pipelines.

TLS stripping attacks expose sensitive web traffic by forcing secure HTTPS connections to fall back to unencrypted HTTP. At present, protection against these attacks relies on website operators explicitly opting into security by deploying mechanisms such as HTTP Strict Transport Security (HSTS) headers. These mechanisms have significant limitations: some are weak or difficult to configure, which raises the risk of misconfiguration and reduces practical adoption; others violate HTTP backward compatibility; at least one can even be abused to enable unintended user tracking. We introduce HSTS-Enforced, a mechanism that eliminates the remaining attack surface for TLS stripping while still allowing operators to securely specify that their websites need to be accessed over HTTP when necessary, thereby maintaining accessibility. To achieve this, we flip the current opt-in security model to an opt-out model: all connections default to HTTPS, and operators can explicitly opt out if their websites require HTTP using so-called HTTP-Required indicators. We propose two such HTTP-Required indicators: a new DNS record and an HTTP-Required Preload list. We evaluate HSTS-Enforced under multiple deployment scenarios, demonstrating that it blocks all practical TLS stripping attempts while maintaining compatibility for sites that require HTTP - without introducing overhead in the typical case. Finally, we outline a practical transition path to accelerate global adoption.