Papers

Synthetic insider threat benchmarks face a consistency problem: corpora generated without an external factual constraint cannot rule out cross-artifact contradictions. The CERT dataset – the field's canonical benchmark – is also static, lacks cross-surface correlation scenarios, and predates the LLM era. We present OrgForge-IT, a verifiable synthetic benchmark in which a deterministic simulation engine maintains ground truth and language models generate only surface prose, making cross-artifact consistency an architectural guarantee. The corpus spans 51 simulated days, 2,904 telemetry records at a 96.4% noise rate, and four detection scenarios designed to defeat single-surface and single-day triage strategies across three threat classes and eight injectable behaviors. A ten-model leaderboard reveals several findings: (1) triage and verdict accuracy dissociate - eight models achieve identical triage F1=0.80 yet split between verdict F1=1.0 and 0.80; (2) baseline false-positive rate is a necessary companion to verdict F1, with models at identical verdict accuracy differing by two orders of magnitude on triage noise; (3) victim attribution in the vishing scenario separates tiers - Tier A models exonerate the compromised account holder while Tier B models detect the attack but misclassify the victim; (4) rigid multi-signal thresholds structurally exclude single-surface negligent insiders, demonstrating the necessity of parallel, threat-class-specific triage pipelines; and (5) agentic software-engineering training acts as a force multiplier for multi-day temporal correlation, but only when paired with frontier-level parameter scale. Finally, prompt sensitivity analysis reveals that unstructured prompts induce vocabulary hallucination, motivating a two-track scoring framework separating prompt adherence from reasoning capability. OrgForge-IT is open source under the MIT license.

The Model Context Protocol (MCP) has rapidly emerged as a universal standard for connecting AI assistants to external tools and data sources. While MCP simplifies integration between AI applications and various services, it introduces significant security vulnerabilities, particularly on the client side. In this work we conduct threat modelings of MCP implementations using STRIDE (Spoofing, Tampering, Repudiation, Information Disclosure, Denial of Service, Elevation of Privilege) and DREAD (Damage, Reproducibility, Exploitability, Affected Users, Discoverability) frameworks across five key components: (1) MCP Host and Client, (2) LLM, (3) MCP Server, (4) External Data Stores, and (5) Authorization Server. This comprehensive analysis reveals tool poisoning-where malicious instructions are embedded in tool metadata-as the most prevalent and impactful client-side vulnerability. We therefore focus our empirical evaluation on this critical attack vector, providing a systematic comparison of how seven major MCP clients validate and defend against tool poisoning attacks. Our analysis reveals significant security issues with most tested clients due to insufficient static validation and parameter visibility. We propose a multi-layered defense strategy encompassing static metadata analysis, model decision path tracking, behavioral anomaly detection, and user transparency mechanisms. This research addresses a critical gap in MCP security, which has primarily focused on server-side vulnerabilities, and provides actionable recommendations and mitigation strategies for securing AI agent ecosystems.

Cryptographic migration driven by algorithm deprecation, regulatory change, and post-quantum readiness requires more than an inventory of cryptographic assets. Existing Cryptographic Bills of Materials (CBOMs) are typically tool- or inventory-derived. They lack architectural intent, rationale, and security context, limiting their usefulness for migration planning. This paper introduces Security-Aware Architecture Tradeoff Analysis Method (SATAM), a security-aware adaptation of scenario-based architecture evaluation that derives an architecture-grounded, context-sensitive CBOM. SATAM integrates established approaches: ATAM, arc42, STRIDE, ADR, and CARAF. These are included to identify and analyze security-relevant cryptographic decision points and document them as explicit architectural decisions. These artifacts are used to annotate CBOM entries with architectural context, security intent, and migration-critical metadata using CycloneDX-compatible extensions. Following a Design Science Research approach, the paper presents the method design, a conceptual traceability model, and an illustrative application. The results demonstrate that architecture-derived CBOMs capture migration-relevant context that is typically absent from inventory-based approaches. Thereby, SATAM improves availability of information required for informed cryptographic migration planning and long-term cryptographic agility.

We present mmFHE, the first system that executes the entire cloud-side mmWave sensing pipeline including the DSP and ML inference under fully homomorphic encryption (FHE). mmFHE encrypts range profiles on an edge device after lightweight plaintext preprocessing and executes the entire mmWave signal-processing and ML inference pipeline homomorphically on a semi-honest cloud that operates exclusively on ciphertexts. At the core of mmFHE is a library of seven composable, data-oblivious FHE kernels that replace standard DSP routines with fixed arithmetic circuits for different application-specific pipelines. We demonstrate this approach on two representative tasks: vital-sign monitoring and gesture recognition. We formally prove two cryptographic guarantees for any pipeline assembled from this library: input privacy and data obliviousness. These guarantees effectively neutralize various supervised and unsupervised privacy attacks on raw data, including re-identification and data-dependent privacy leakage. Evaluation on three public radar datasets shows that encryption introduces negligible error versus the plaintext pipeline, with 84.5% gesture accuracy (vs. 84.7%). End-to-end cloud GPU latency is 1.21 s per 10 s vital-sign window and 5.76 s per 3 s gesture window. These results establish the initial feasibility of end-to-end mmWave sensing under FHE on commodity hardware.

With the rapid growth of interconnected devices, accurately detecting malicious activities in network traffic has become increasingly challenging. Most existing deep learning-based intrusion detection systems treat network flows as independent instances, thereby failing to exploit the relational dependencies inherent in network communications. To address this limitation, we propose Q-AGNN, a Quantum-Enhanced Attentive Graph Neural Network for intrusion detection, where network flows are modeled as nodes and edges represent similarity relationships. Q-AGNN leverages parameterized quantum circuits (PQCs) to encode multi-hop neighborhood information into a high-dimensional latent space, inducing a bounded quantum feature map that implements a second-order polynomial graph filter in a quantum-induced Hilbert space. An attention mechanism is subsequently applied to adaptively weight the quantum-enhanced embeddings, allowing the model to focus on the most influential nodes contributing to anomalous behavior. Extensive experiments conducted on four benchmark intrusion detection datasets demonstrate that Q-AGNN achieves competitive or superior detection performance compared to state-of-the-art graph-based methods, while consistently maintaining low false positive rates under hardware-calibrated noise conditions. Moreover, we also executed the Q-AGNN framework on actual IBM quantum hardware to demonstrate the practical operability of the proposed pipeline under real NISQ conditions. These results highlight the effectiveness of integrating quantum-enhanced representations with attention mechanisms for graph-based intrusion detection and underscore the potential of hybrid quantum-classical learning frameworks in cybersecurity applications.

A Large Language Model (LLM) as judge evaluates the quality of victim Machine Learning (ML) models, specifically LLMs, by analyzing their outputs. An LLM as judge is the combination of one model and one specifically engineered judge prompt that contains the criteria for the analysis. The resulting automation of the analysis scales up the complex evaluation of the victim models' free-form text outputs by faster and more consistent judgments compared to human reviewers. Thus, quality and security assessments of LLMs can cover a wide range of the victim models' use cases. Being a comparably new technique, LLMs as judges lack a thorough investigation for their reliability and agreement to human judgment. Our work evaluates the applicability of LLMs as automated quality assessors of victim LLMs. We test the efficacy of 37 differently sized conversational LLMs in combination with 5 different judge prompts, the concept of a second-level judge, and 5 models fine-tuned for the task as assessors. As assessment objective, we curate datasets for eight different categories of judgment tasks and the corresponding ground-truth labels based on human assessments. Our empirical results show a high correlation of LLMs as judges with human assessments, when combined with a suitable prompt, in particular for GPT-4o, several open-source models with $\geqslant$ 32B parameters, and a few smaller models like Qwen2.5 14B.

The use of Internet of Things (IoT) devices is growing at a rapid rate. While much of this growth is consumer devices, IoT devices are also commonly found in corporate and industrial environments, as well. These devices can be organization-owned and managed by an information technology unit, deployed organizationally without the knowledge and involvement of technology staff or brought in to the corporate environment by user-owners. In each case, these devices may have access to corporate networks and data and are, thus, important to consider as part of organizational cybersecurity risk assessment. Despite the prevalence of these devices, there is little literature about how to audit their security. This paper presents a risk-based auditing framework which can be used by both internal and external auditors, of any experience level and in any industry, to assess IoT devices.

We present TALUS, a threshold signing protocol for ML-DSA (FIPS 204) that supports an arbitrary number of parties, emits signatures accepted by any unmodified ML-DSA verifier, and reduces, for a bounded number of signatures per key, to ML-DSA's own assumptions (Module-LWE and SelfTargetMSIS). Its core is the Boundary Clearance Condition (BCC): for a constant fraction of nonces (31.7% at ML-DSA-65), the secret vector s2 provably cannot cross a rounding boundary, so the s2-dependent rejection check can be enforced offline on preprocessed nonces. This removes the interaction that forces multi-round signing in prior schemes, yielding a TEE-assisted one-round profile and a fully distributed honest-majority MPC profile with two online rounds, both built on a Carry Elimination Framework that computes the shared commitment w1 = HighBits(Ay) on secret shares, a step we show is unavoidable. We prove a lower bound: any FIPS-204-exact threshold scheme revealing a summed (Irwin-Hall) nonce admits an efficient key-recovery attack after about 2^30 signatures under one key (at ML-DSA-65), so no q_s-independent unforgeability is possible for this class. Complementing it, a q_s-bounded analysis certifies over 105 bits at the operational signing cap for the external view, with insider ceilings disclosed per view and a mandatory key rotation capping each key's signing lifetime. A Rust implementation across all three FIPS 204 levels demonstrates practicality.

In Shamir's secret sharing scheme, all participants possess equal privileges. However, in many practical scenarios, it is often necessary to assign different levels of authority to different participants. To address this requirement, Hierarchical Secret Sharing (HSS) schemes were developed, which partitioned all participants into multiple subsets and assigned a distinct privilege level to each. Existing Chinese Remainder Theorem (CRT)-based HSS schemes benefit from flexible share sizes, but either exhibit security flaws or have an information rate less than $\frac{1}{2}$. In this work, we propose a disjunctive HSS scheme and a conjunctive HSS scheme by using the CRT for integer ring and one-way functions. Both schemes are asymptotically ideal and are proven to be secure.

Conjunctive Hierarchical Secret Sharing (CHSS) is a type of secret sharing that divides participants into multiple distinct hierarchical levels, with each level having a specific threshold. An authorized subset must simultaneously meet the threshold of all levels. Existing Chinese Remainder Theorem (CRT)-based CHSS schemes either have security vulnerabilities or have an information rate lower than $\frac{1}{2}$. In this work, we utilize the CRT for polynomial ring and one-way functions to construct an asymptotically perfect CHSS scheme. It has computational security, and permits flexible share sizes. Notably, when all shares are of equal size, our scheme is an asymptotically ideal CHSS scheme with an information rate one.

Large language models are becoming pervasive core components in many real-world applications. As a consequence, security alignment represents a critical requirement for their safe deployment. Although previous related works focused primarily on model architectures and alignment methodologies, these approaches alone cannot ensure the complete elimination of harmful generations. This concern is reinforced by the growing body of scientific literature showing that attacks, such as jailbreaking and prompt injection, can bypass existing security alignment mechanisms. As a consequence, additional security strategies are needed both to provide qualitative feedback on the robustness of the obtained security alignment at the training stage, and to create an “ultimate” defense layer to block unsafe outputs possibly produced by deployed models. To provide a contribution in this scenario, this paper introduces SecureBreak, a safety-oriented dataset designed to support the development of AI-driven solutions for detecting harmful LLM outputs caused by residual weaknesses in security alignment. The dataset is highly reliable due to careful manual annotation, where labels are assigned conservatively to ensure safety. It performs well in detecting unsafe content across multiple risk categories. Tests with pre-trained LLMs show improved results after fine-tuning on SecureBreak. Overall, the dataset is useful both for post-generation safety filtering and for guiding further model alignment and security improvements.

New technologies, such as blockchain, are designed to address various system weaknesses, particularly those related to security. Blockchain can enhance numerous aspects of traditional banking systems by transforming them into digital, immutable, secure, and anonymous ledger. This paper proposes a new banking application ALBank, which is based on blockchain and smart contract technologies. Its functionality relies on invoking functions within smart contracts deployed on the Ethereum blockchain. This approach enables decentralization and enhances both security and trust. In this context, the paper first presents a critical analysis of existing research on blockchain and traditional banking systems, with a focus on their respective challenges. It then examines the Know Your Customer (KYC) process and its various models. Finally, it introduces the design and development of ALBank, a decentralized banking application built on the Ethereum blockchain using smart contracts. The results show that the integration of blockchain and smart contracts effectively addresses key issues in traditional banking systems, including centralization, inefficiency, and security vulnerabilities by storing critical data on a decentralized, immutable ledger, managing processes autonomously, and making transactions transparent to all users.

Modern democracies face an existential crisis of waning public trust in election results. While End-to-End Verifiable (E2E-V) voting systems promise mathematically secure elections, their reliance on complex cryptography creates a “black box” that forces blind trust in opaque software or external experts, ultimately failing to build genuine public confidence. To solve this, we introduce the concept of Software-Free Verification (SFV) – a standard requiring that voters can independently verify election integrity without relying on any software. We propose a practical, non-cryptographic in-booth voting scheme that achieves SFV for national-scale elections. Our approach leverages a public bulletin board of randomized (Pseudonym, Candidate) pairs, where a mechanically generated pseudonym is hidden among real decoy votes on a physical receipt. Our scheme empowers citizens to audit the election using only basic arithmetic via a hierarchical Public Ledger, while anchoring the overall digital tally to physical evidence and Risk-Limiting Audits (RLAs) to guarantee systemic integrity. The result is a system that bridges the gap between mathematical security and public transparency, offering a viable blueprint for restoring trust in democratic infrastructure.

This paper emphasizes the critical role of interoperability in enabling efficient and secure communication for the fragmented distributed ledger ecosystem, particularly within on-chain finance. The purpose of this study is to streamline and accelerate empirical research on the intersection of cross-chain interoperability solutions and their impact within on-chain finance. The analysis examines the relationship between financial use and interoperability while comparing the properties of novel cross-chain interoperability protocols (LayerZero, Wormhole, Connext, Chainlink Cross-Chain Interoperability Protocol, Circle Cross-chain Transfer Protocol, Hop Protocol, Across, Polkadot, and Cosmos), focusing on their design, mechanisms, consensus, and limitations. To encourage further empirical study, the paper proposes a set of network metrics and sample statistical models and provides a framework for evaluating the performance and financial implications of interoperability solutions.

Smart homes are increasingly targeted by cyberattacks, yet residents often lack guidance when incidents occur. Since affected residents are likely to seek help from trustworthy sources, this paper asks: What actionable cybersecurity guidance do governments provide to smart home users whose systems have been compromised? To answer this question, we conduct an exploratory, user-centered review of governmental cybersecurity guidance for smart homes across eleven countries to identify and characterize the types of guidance governments provide and to systematize their content. Using a standardized search and screening process, we derive three emergent clusters: incident reporting, general security recommendations, and incident response. Our findings show that governments provide abundant general security advice and accessible reporting channels, but structured incident response guidance tailored to smart homes is rare. Only two sources offer step-by-step recovery guidance for non-expert users, highlighting a gap between preventive advice and post-incident support.

Multimodal Large Language Models (MLLMs) extend text-only LLMs with visual reasoning, but also introduce new safety failure modes under visually grounded instructions. We study comic-template jailbreaks that embed harmful goals inside simple three-panel visual narratives and prompt the model to role-play and "complete the comic." Building on JailbreakBench and JailbreakV, we introduce ComicJailbreak, a comic-based jailbreak benchmark with 1,167 attack instances spanning 10 harm categories and 5 task setups. Across 15 state-of-the-art MLLMs (six commercial and nine open-source), comic-based attacks achieve success rates comparable to strong rule-based jailbreaks and substantially outperform plain-text and random-image baselines, with ensemble success rates exceeding 90% on several commercial models. Then, with the existing defense methodologies, we show that these methods are effective against the harmful comics, they will induce a high refusal rate when prompted with benign prompts. Finally, using automatic judging and targeted human evaluation, we show that current safety evaluators can be unreliable on sensitive but non-harmful content. Our findings highlight the need for safety alignment robust to narrative-driven multimodal jailbreaks.

The present work investigates a type of morphisms between encryption schemes, called bridges. By associating an encryption scheme to every such bridge, we define and examine their security. Inspired by the bootstrapping procedure used by Gentry to produce fully homomorphic encryption schemes, we exhibit a general recipe for the construction of bridges. Our main theorem asserts that the security of a bridge reduces to the security of the first encryption scheme together with a technical additional assumption.

Retrieval-Augmented Generation (RAG) significantly mitigates the hallucinations and domain knowledge deficiency in large language models by incorporating external knowledge bases. However, the multi-module architecture of RAG introduces complex system-level security vulnerabilities. Guided by the RAG workflow, this paper analyzes the underlying vulnerability mechanisms and systematically categorizes core threat vectors such as data poisoning, adversarial attacks, and membership inference attacks. Based on this threat assessment, we construct a taxonomy of RAG defense technologies from a dual perspective encompassing both input and output stages. The input-side analysis reviews data protection mechanisms including dynamic access control, homomorphic encryption retrieval, and adversarial pre-filtering. The output-side examination summarizes advanced leakage prevention techniques such as federated learning isolation, differential privacy perturbation, and lightweight data sanitization. To establish a unified benchmark for future experimental design, we consolidate authoritative test datasets, security standards, and evaluation frameworks. To the best of our knowledge, this paper presents the first end-to-end survey dedicated to the security of RAG systems. Distinct from existing literature that isolates specific vulnerabilities, we systematically map the entire pipeline-providing a unified analysis of threat models, defense mechanisms, and evaluation benchmarks. By enabling deep insights into potential risks, this work seeks to foster the development of highly robust and trustworthy next-generation RAG systems.

Phishing attacks remain a persistent cybersecurity threat, and the widespread adoption of TLS certificates has unintentionally enabled malicious websites to appear trustworthy to users. This study examines whether certificate metadata and domain characteristics can help distinguish phishing domains from benign domains within the Danish .dk namespace. A dataset was constructed by combining registry information from Punktum dk with phishing reports and popularity rankings from external sources. TLS certificate attributes were collected using Netlas, while additional domain-based features were derived from DNS records and lexical analysis of domain names. The analysis compares phishing, popular, and less frequently visited domains across several feature categories, including Certificate Authorities (CAs), validity periods, missing certificate fields, SAN structure, registrant geography, hosting providers, and lexical properties of domain names. The results indicate that several features show observable differences between phishing and highly popular domains. However, phishing domains often resemble less popular domains, resulting in substantial overlap across many characteristics. Consequently, no individual feature provides a reliable standalone indicator of phishing activity within the Danish namespace. The findings suggest that certificate and domain attributes may still contribute to detection when combined, while also highlighting the limitations of relying on individual indicators in isolation. This work provides an empirical overview of phishing-related infrastructure patterns in the Danish .dk ecosystem and offers insights that may inform future phishing detection approaches.

Prompt injection is listed as the number-one vulnerability class in the OWASP Top 10 for LLM Applications that can subvert LLM guardrails, disclose sensitive data, and trigger unauthorized tool use. Developers are rapidly adopting AI-assisted development tools built on the Model Context Protocol (MCP). However, their convenience comes with security risks, especially prompt-injection attacks delivered via tool-poisoning vectors. While prior research has studied prompt injection in LLMs, the security posture of real-world MCP clients remains underexplored. We present the first empirical analysis of prompt injection with the tool-poisoning vulnerability across seven widely used MCP clients: Claude Desktop, Claude Code, Cursor, Cline, Continue, Gemini CLI, and Langflow. We identify their detection and mitigation mechanisms, as well as the coverage of security features, including static validation, parameter visibility, injection detection, user warnings, execution sandboxing, and audit logging. Our evaluation reveals significant disparities. While some clients, such as Claude Desktop, implement strong guardrails, others, such as Cursor, exhibit high susceptibility to cross-tool poisoning, hidden parameter exploitation, and unauthorized tool invocation. We further provide actionable guidance for MCP implementers and the software engineering community seeking to build secure AI-assisted development workflows.

The Model Context Protocol (MCP) has emerged as a standard for connecting Large Language Models (LLMs) to external tools and data. However, MCP servers often expose privileged capabilities, such as file system access, network requests, and command execution that can be exploited if not properly secured. We present mcp-sec-audit, an extensible security assessment toolkit designed specifically for MCP servers. It implements static pattern matching for Python-based MCP servers and dynamic sandboxed fuzzing and monitoring via Docker and eBPF. The tool detects risky capabilities through configurable rule-based analysis and provides mitigation recommendations.

The rapid expansion of the Internet of Things (IoT) and its integration with backbone networks have heightened the risk of security breaches. Traditional centralized approaches to anomaly detection, which require transferring large volumes of data to central servers, suffer from privacy, scalability, and latency limitations. This paper proposes a lightweight autoencoder-based anomaly detection framework designed for deployment on resource-constrained edge devices, enabling real-time detection while minimizing data transfer and preserving privacy. Federated learning is employed to train models collaboratively across distributed devices, where local training occurs on edge nodes and only model weights are aggregated at a central server. A real-world IoT testbed using Raspberry Pi sensor nodes was developed to collect normal and attack traffic data. The proposed federated anomaly detection system, implemented and evaluated on the testbed, demonstrates its effectiveness in accurately identifying network attacks. The communication overhead was reduced significantly while achieving comparable performance to the centralized method.

Developers rely on online tutorials to learn web application security, but tutorial quality varies. We reviewed 132 free security tutorials to examine topic coverage, authorship, and technical depth. Our analysis shows that most tutorials come from vendors and emphasize high-level explanations over concrete implementation guidance. Few tutorials provide complete runnable code examples or direct links to authoritative security resources such as the Open Web Application Security Project (OWASP), Common Weakness Enumeration (CWE), or Common Vulnerabilities and Exposures (CVE). We found that two visible signals help identify more useful tutorials: the presence of runnable code and direct links to official resources. These signals can help developers distinguish broad awareness material from tutorials that better support secure implementation.

To comply with data protection regulations such as the EU General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA), websites widely deploy cookie consent banners to collect users' privacy preferences. In practice, however, these interfaces often embed dark patterns that undermine informed and freely given consent. As regulatory scrutiny increases, such patterns have not disappeared but have evolved into subtler and more legally ambiguous forms, making existing detection approaches outdated. We present UMBRA, a consent management platform (CMP)-agnostic system that detects both previously studied patterns (DP1-DP10) and nine newly evolved patterns (DP11-DP19) targeting information disclosure, consent revocation, and legal ambiguity, including pay-to-opt-out schemes, revocation barriers, and fake opt-outs. UMBRA combines text analysis, visual heuristics, interaction tracing, and cookie-state monitoring to capture multi-step consent flows missed by prior tools. We evaluate UMBRA on a manually annotated ground-truth dataset and achieve 99% detection accuracy. We further conduct a large-scale compliance-oriented measurement across 14,000 websites spanning the EU, the US, and top-ranked global domains. Our results show that evolved dark patterns are pervasive: revocation is often obstructed, cookies are set before consent or despite explicit rejection, and opt-out interfaces often fail to prevent third-party tracking. On sites with revocation barriers, cookies increase by 25% on average, and many use insecure attributes that increase exposure to attacks such as XSS and CSRF. Overall, our findings provide evidence of systematic non-compliance and show how evolving consent manipulation erodes user autonomy while amplifying privacy and security risks.

In this work, we identify a set of side-channels in our Confidential Federated Compute platform that a hypothetical insider could exploit to circumvent differential privacy (DP) guarantees. We show how DP can mitigate two of the side-channels, one of which has been implemented in our open-source library.

Deterministic pre-execution safety gates evaluate whether individual agent actions are compatible with their assigned roles. While effective at per-action authorization, these systems are structurally blind to distributed attacks that decompose harmful intent across multiple individually-compliant steps. This paper introduces Session Risk Memory (SRM), a lightweight deterministic module that extends stateless execution gates with trajectory-level authorization. SRM maintains a compact semantic centroid representing the evolving behavioral profile of an agent session and accumulates a risk signal through exponential moving average over baseline-subtracted gate outputs. It operates on the same semantic vector representation as the underlying gate, requiring no additional model components, training, or probabilistic inference. We evaluate SRM on a multi-turn benchmark of 80 sessions containing slow-burn exfiltration, gradual privilege escalation, and compliance drift scenarios. Results show that ILION+SRM achieves F1 = 1.0000 with 0% false positive rate, compared to stateless ILION at F1 = 0.9756 with 5% FPR, while maintaining 100% detection rate for both systems. Critically, SRM eliminates all false positives with a per-turn overhead under 250 microseconds. The framework introduces a conceptual distinction between spatial authorization consistency (evaluated per action) and temporal authorization consistency (evaluated over trajectory), providing a principled basis for session-level safety in agentic systems.

As large language models are deployed as autonomous agents with tool execution privileges, a critical assumption underpins their security architecture: that model errors are detectable at runtime. We present empirical evidence that this assumption fails for two of three instruction-following models evaluable for conflict detection. We introduce governability – the degree to which a model's errors are detectable before output commitment and correctable once detected – and demonstrate it varies dramatically across models. In six models across twelve reasoning domains, two of three instruction-following models exhibited silent commitment failure: confident, fluent, incorrect output with zero warning signal. The remaining model produced a detectable conflict signal 57 tokens before commitment under greedy decoding. We show benchmark accuracy does not predict governability, correction capacity varies independently of detection, and identical governance scaffolds produce opposite effects across models. A 2x2 experiment shows a 52x difference in spike ratio between architectures but only +/-0.32x variation from fine-tuning, suggesting governability is fixed at pretraining. We propose a Detection and Correction Matrix classifying model-task combinations into four regimes: Governable, Monitor Only, Steer Blind, and Ungovernable.

Adversarial-example-based fingerprinting approaches, which leverage the decision boundary characteristics of deep neural networks (DNNs) to craft fingerprints, have proven effective for model ownership protection. However, a fundamental challenge remains unresolved: how far a fingerprint should be placed from the decision boundary to simultaneously satisfy two essential properties, i.e., robustness and uniqueness, for effective and reliable ownership protection. Despite the importance of the fingerprint-to-boundary distance, existing works lack a theoretical solution and instead rely on empirical heuristics, which may violate either robustness or uniqueness properties. We propose AnaFP, an analytical fingerprinting scheme that constructs fingerprints under theoretical guidance. Specifically, we formulate fingerprint generation as controlling the fingerprint-to-boundary distance through a tunable stretch factor. To ensure both robustness and uniqueness, we mathematically formalize these properties that determine the lower and upper bounds of the stretch factor. These bounds jointly define an admissible interval within which the stretch factor must lie, thereby establishing a theoretical connection between the two constraints and the fingerprint-to-boundary distance. To enable practical fingerprint generation, we approximate the original (infinite) sets of pirated and independently trained models using two finite surrogate model pools and employ a quantile-based relaxation strategy to relax the derived bounds. Due to the circular dependency between the lower bound and the stretch factor, we apply grid search over the admissible interval to determine the most feasible stretch factor. Extensive experimental results show that AnaFP consistently outperforms prior methods, achieving effective ownership verification across diverse model architectures and model modification attacks.

Advanced Persistent Threats (APTs) are stealthy, multi-stage attacks that require adaptive and timely defense. While deep reinforcement learning (DRL) enables autonomous cyber defense, its decisions are often opaque and difficult to trust in operational environments. This paper presents DeepXplain, an explainable DRL framework for stage-aware APT defense. Building on our prior DeepStage model, DeepXplain integrates provenance-based graph learning, temporal stage estimation, and a unified XAI pipeline that provides structural, temporal, and policy-level explanations. Unlike post-hoc methods, explanation signals are incorporated directly into policy optimization through evidence alignment and confidence-aware reward shaping. To the best of our knowledge, DeepXplain is the first framework to integrate explanation signals into reinforcement learning for APT defense. Experiments in a realistic enterprise testbed show improvements in stage-weighted F1-score (0.887 to 0.915) and success rate (84.7% to 89.6%), along with higher explanation confidence (0.86), improved fidelity (0.79), and more compact explanations (0.31). These results demonstrate enhanced effectiveness and trustworthiness of autonomous cyber defense.

At S&P 2023, Puschner et al. made a valuable dataset for hardware Trojan detection research publicly available. It contains a complete set of Scanning Electron Microscope (SEM) images of four different digital Integrated Circuits (ICs) fabricated at progressively smaller semiconductor technology nodes. Puschner et al. reported preliminary evidence that feature sizes affect Trojan detection performance, but they were unable to disentangle effects caused by insertion strategies or by degrading image quality from those intrinsic to the underlying standard cell libraries. Distinguishing those causes, however, is crucial to understand whether improved tooling (e.g., higher resolution imaging equipment) can remove the observed technology bias, or whether susceptibility to stealthy hardware Trojans is indeed an inherent property of a cell library. In this work, we dive deep into the S&P 2023 dataset to answer these questions. We devise alternative metrics to those of Puschner et al., in order to assess and compare the potential susceptibility of standard cell libraries more meaningfully. We find clear differences between the evaluated process nodes. However, in all cases we identify cells that implement distinct logic functions yet are visually indistinguishable in backside SEM images. We exploit this property to construct stealthy, standard-cell-based hardware Trojans and present a concrete case study: a privilege-escalation backdoor in an Ibex RISCV core. Our results demonstrate that cell libraries can - and should - be evaluated for their potential "Trojanizability", and we recommend practical defenses.