Papers

Digital image steganography requires a careful trade-off among payload capacity, visual fidelity, and statistical undetectability. Fixed-depth least significant bit embedding remains attractive because of its simplicity and high capacity, but it modifies smooth and textured regions uniformly, thereby increasing distortion and detectability in statistically sensitive areas. This paper presents an adaptive steganographic framework that combines a Mamdanitype fuzzy inference system with modern authenticated encryption. The proposed method determines a pixel-wise embedding depth from 1 to 3 bits using local entropy, edge magnitude, and payload pressure as linguistic inputs. To preserve encoder-decoder synchronization, the same feature maps are computed from lower-bit-stripped images, making the adaptive control mechanism invariant to the least significant modifications introduced during embedding. A cryptographic layer based on Argon2id and AES-256-GCM protects payload confidentiality and integrity independently of steganographic concealment.

With the widespread deployment of deep-learning-based speech models in security-critical applications, backdoor attacks have emerged as a serious threat: an adversary who poisons a small fraction of training data can implant a hidden trigger that controls the model's output while preserving normal behavior on clean inputs. Existing inference-time defenses are not well suited to the audio domain, as they either rely on trigger over-robustness assumptions that fail on transformation-based and semantic triggers, or depend on properties specific to image or text modalities. In this paper, we propose STEP (Stability-based Trigger Exposure Profiling), a black-box, retraining-free backdoor detector that operates under hard-label-only access. Its core idea is to exploit a characteristic dual anomaly of backdoor triggers: anomalous label stability under semantic-breaking perturbations, and anomalous label fragility under semantic-preserving perturbations. STEP profiles each test sample with two complementary perturbation branches that target these two properties respectively, scores the resulting stability features with one-class anomaly detectors trained on benign references, and fuses the two scores via unsupervised weighting. Extensive experiments across seven backdoor attacks show that STEP achieves an average AUROC of 97.92% and EER of 4.54%, substantially outperforming state-of-the-art baselines, and generalizes across model architectures, speech tasks, an open-set verification scenario, and over-the-air physical-world settings.

We introduce list privacy amplification (LPA), a relaxation of the final step of quantum key distribution (QKD) in which Alice and Bob extract a list of $L$ candidate keys from a raw string correlated with an eavesdropper Eve, with the guarantee that at least one key is perfectly secret while Eve cannot identify which. This parallels list decoding in error-correcting codes: relaxing unique decoding to list decoding increases the decoding radius; analogously, list extraction increases achievable key length beyond the standard quantum leftover hash lemma (QLHL). Within the abstract cryptography framework, we formalise LPA and prove the Quantum List Leftover Hash Lemma (QLLHL): an $L$-list of $\ell$-bit keys can be extracted from an $n$-bit source with smooth min-entropy $k$ iff \[ \ell \le k + \log L - 2\log(1/\epsilon) - 3, \] yielding a tight additive $\log L$ gain over QLHL. This gain arises because the index of the secure key is chosen after hashing and hidden from Eve, effectively contributing $\log L$ bits of entropy. Applying QLLHL to BB84-type QKD, a list size $L = 2^{\alpha n'}$ increases the tolerable phase-error threshold from $h^{-1}(1 - h(e_b))$ to $h^{-1}(1 - h(e_b) + \alpha)$, exceeding the standard $\approx 11\%$ bound for any $\alpha > 0$. We prove tightness via a matching intercept-resend attack, establish composability with Wegman–Carter authentication, and present two constructions: a polynomial inner-product hash over $\mathbb{F}_{2^m}$ and a Toeplitz-based variant, running in $O(nL)$ and $O(nL \log n)$ time.

Code smoothing is a phenomenon in which an error distribution makes a code statistically close to the uniform distribution over the ambient space. This closeness is measured by the total variation distance. Recently, Debris-Alazard et al.\ introduced a smoothing bound, which is an upper bound on this total variation distance. Although the smoothing bound evaluates how the error distribution smooths a code, this bound applies only to linear codes. In this paper, we generalize this bound to not only linear codes but also specific non-linear codes. While the smoothing bound in previous work was obtained by Fourier analysis over finite abelian groups, we derive this bound using a graph-theoretic approach. To derive the smoothing bound, we consider code smoothing as the mixing of random walks on a specific graph, and use the concept of equitable partitions, which is well-studied in graph theory.

We present an experience report on YouTube-Synch, a production system that replicates content from 10,000+ creator-authorized YouTube channels to a blockchain-based decentralized platform. Although replication is authorized, the system must still defeat YouTube's anti-automation defenses - API quota restrictions (10,000 units/day), IP-based rate limiting, behavioral bot detection, and OAuth token lifecycle policies - which do not distinguish authorized bulk access from abuse. Our central observation is a coupled-defense phenomenon: these protection layers are not independent, so circumventing one (e.g., API quotas) silently activates another (e.g., OAuth token expiration), producing cascading, delayed failures. We ground this in three production incidents with concrete impact - 28 duplicate on-chain objects from a database throughput failure, 10,000+ channels lost to a single OAuth mass-expiration, and 719 daily errors from queue pollution - observed over 15 releases and 3.5 years of operation. We further argue, from the system's own concurrency and rate parameters alone, that detection-driven anti-bot measures impose a download-bound throughput ceiling on the order of 10^3 videos/day per instance - at or below the steady-state demand of 10,000 channels. This analysis shows why priority-based triage and horizontal scaling are structural necessities rather than optimizations, and explains the survivability-over-throughput tradeoff that drove a 25x reduction in download concurrency. We detail the resulting architecture - a four-stage DAG pipeline, a Write-Ahead-Log fault-tolerance model with cross-system state reconciliation, and a trust-minimized ownership-verification protocol that eliminates OAuth - and distill design principles that generalize to extraction against other heavily-defended centralized platforms.

The Model Context Protocol (MCP) introduces a structurally distinct attack surface that existing threat frameworks, designed for traditional software systems or generic LLM deployments, do not adequately cover. This paper presents MCP-38, a protocol-specific threat taxonomy consisting of 38 threat categories (MCP-01 through MCP-38). The taxonomy was derived through a systematic four-phase methodology: protocol decomposition, multi-framework cross-mapping, real-world incident synthesis, and remediation-surface categorization. Each category is mapped to STRIDE, OWASP Top 10 for LLM Applications (2025, LLM01–LLM10), and the OWASP Top 10 for Agentic Applications (2026, ASI01–ASI10). MCP-38 addresses critical threats arising from MCP's semantic attack surface (tool description poisoning, indirect prompt injection, parasitic tool chaining, and dynamic trust violations), none of which are adequately captured by prior work. MCP-38 provides the definitional and empirical foundation for automated threat intelligence platforms.

Tool-using automation systems, from scripts and CI bots to agentic assistants, fail in recurring patterns. Common failures include unsafe side effects, invalid arguments, uncontrolled retries, and leakage of sensitive outputs. Many mitigations are model-centric and prompt-dependent, so they are brittle and do not generalize to non-LLM callers. We present Policy-First Tooling, a model-agnostic permission layer that mediates tool invocation through explicit constraints, risk-aware gating, recovery controls, and auditable explanations. The paper contributes a compact policy DSL, a runtime enforcement architecture with actionable rationale and fix hints, and a reproducible benchmark based on trace replay with controlled fault and misuse injection. In 225 controlled runs across five policy packs and three fault profiles, stricter packs improve violation prevention from 0.000 in P0 to 0.681 in P4, while task success drops from 0.356 to 0.067. Retry amplification decreases from 3.774 in P0 to 1.378 in P4, and leakage recall reaches 0.875 under injected secret outputs. These results make safety to utility trade-offs explicit and measurable.

Large language models (LLMs) and AI agents are increasingly integrated into enterprise systems to access internal databases and generate context-aware responses. While such integration improves productivity and decision support, the model outputs may inadvertently reveal sensitive information. Although many prior efforts focus on protecting the privacy of user prompts, relatively few studies consider privacy risks from the enterprise data perspective. Hence, this paper develops a probabilistic framework for analyzing privacy leakage in AI agents based on differential privacy. We model response generation as a stochastic mechanism that maps prompts and datasets to distributions over token sequences. Within this framework, we introduce token-level and message-level differential privacy and derive privacy bounds that relate privacy leakage to generation parameters such as temperature and message length. We further formulate a privacy-utility design problem that characterizes optimal temperature selection.

Hardware serves as the root of trust in modern computing systems, making Hardware Reverse Engineering (HRE) essential for security assurance$-$from design verification and supply-chain integrity to vulnerability discovery. We scope HRE to netlist recovery and its subsequent analysis, spanning the three subdomains of Integrated Circuit (IC), Field-Programmable Gate Array (FPGA), and netlist reverse engineering. These subdomains differ in their methodologies, but share core processes and are shaped by common requirements and legal constraints of the same stakeholders. Despite an increasing number of publications, the field lacks a systematic understanding of how these obstacles have stunted the research ecosystem. To address this gap, we present the first large-scale Systematization of Knowledge (SoK) of the HRE workflow, analyzing 187 peer-reviewed publications. Across all three subdomains, we identify eleven concrete technical challenges$-$from a widening gap between academic research and modern semiconductor technology nodes to overly idealized assumptions in netlist analysis$-$and propose actionable directions for each. A retrospective evaluation of all 30 published artifacts reveals that key results could be reproduced for only seven, a mere 4 % of all 187 papers in our corpus, confirming a systemic reproducibility crisis. We trace both the technical and reproducibility challenges to three structural barriers that recur across all subdomains: scarce reusable artifacts, missing benchmarks, and unresolved legal constraints on data sharing and collaboration. Based on these findings, we derive stakeholder-specific recommendations for academia, industry, and government to transition HRE from isolated research silos toward a collaborative discipline capable of assuring increasingly complex, global hardware supply chains.

RISC-V-based Trusted Execution Environments (TEEs) are gaining traction in the automotive and IoT sectors as a foundation for protecting sensitive computations. However, the supporting infrastructure around these TEEs remains immature. In particular, mechanisms for secure enclave updates and migrations - essential for complete enclave lifecycle management - are largely absent from the evolving RISC-V ecosystem. In this paper, we address this limitation by introducing a novel toolkit that enables RISC-V TEEs to support critical aspects of the software development lifecycle. Our toolkit provides broad compatibility with existing and emerging RISC-V TEE implementations (e.g., Keystone and CURE), which are particularly promising for integration in the automotive industry. It extends the Security Monitor (SM) - the trusted firmware layer of RISC-V TEEs - with three modular extensions that enable secure enclave update, secure migration, state continuity, and trusted time. Our implementation demonstrates that the toolkit requires only minimal interface adaptation to accommodate TEE-specific naming conventions. Our evaluation results confirm that our proposal introduces negligible performance overhead: our state continuity solution incurs less than 1.5% overhead, and enclave downtime remains as low as 0.8% for realistic applications with a 1 KB state, which conforms with the requirements of most IoT and automotive applications.

Quantum computing often requires classical data to be supplied to execution environments that may not be fully trusted or isolated. While encryption protects data at rest and in transit, it provides limited protection once computation begins, when classical values are encoded into quantum registers. This paper explores data obfuscation for protecting classical values during quantum computation. To the best of our knowledge, we present the first explicit data obfuscation technique designed to protect classical values during quantum execution. We propose an obfuscation technique that encodes sensitive data into structured quantum representations across multiple registers, avoiding direct exposure while preserving computational usability. Reversible quantum operations and amplitude amplification allow selective recovery of valid encodings without revealing the underlying data. We evaluate the feasibility of the proposed method through simulation and analyze its resource requirements and practical limitations. Our results highlight data obfuscation as a complementary security primitive for quantum computing.

Supervised detection of network attacks has always been a critical part of network intrusion detection systems (NIDS). Nowadays, in a pivotal time for artificial intelligence (AI), with even more sophisticated attacks that utilize advanced techniques, such as generative artificial intelligence (GenAI) and reinforcement learning, it has become a vital component if we wish to protect our personal data, which are scattered across the web. In this paper, we address two tasks, in the first unified multi-modal NIDS dataset, which incorporates flow-level data, packet payload information and temporal contextual features, from the reprocessed CIC-IDS-2017, CIC-IoT-2023, UNSW-NB15 and CIC-DDoS-2019, with the same feature space. In the first task we use machine learning (ML) algorithms, with stratified cross validation, in order to prevent network attacks, with stability and reliability. In the second task we use adversarial learning algorithms to generate synthetic data, compare them with the real ones and evaluate their fidelity, utility and privacy using the SDV framework, f-divergences, distinguishability and non-parametric statistical tests. The findings provide stable ML models for intrusion detection and generative models with high fidelity and utility, by combining the Synthetic Data Vault framework, the TRTS and TSTR tests, with non-parametric statistical tests and f-divergence measures.

LLM agents are becoming increasingly important in the security domain, but leading systems are often closed-source, cloud-based, hard to reproduce or use with sensitive code. This creates a need for small, local models that can perform security tasks under strict resource constraints, though effective methods for developing them remain unexplored. In this paper, we address this gap by proposing a two-stage post-training recipe that turns a small local language model into a security agent. To this end, we focus on Linux privilege escalation as a representative setting to systematically study the training of local models, as the task is both automatically verifiable and requires multi-step interactive reasoning. Using an experimental setup that mitigates data leakage, we post-train a small 4B model in two stages: supervised fine-tuning on traces from procedural privilege-escalation environments, followed by reinforcement learning with verifiable rewards. On a held-out benchmark of 12 Linux privilege-escalation scenarios, supervised fine-tuning doubles the baseline success rate under a tight budget of 20 interaction rounds, and subsequent reinforcement learning training improves our model, PrivEsc-LLM 4B, to 93.3% success, behind only Claude Opus 4.7 at this budget. At the same time, the expected inference cost per successful escalation decreases by more than 80x. Our findings not only show that small local models can be adapted to complex security tasks, but also document the challenges involved, offering guidance for transferring this recipe to other settings.

Federated Learning (FL) enables collaborative model training by sharing model updates instead of raw data, aiming to protect user privacy. However, recent studies reveal that these shared updates can inadvertently leak sensitive training data through gradient inversion attacks (GIAs). Among them, active GIAs are particularly powerful, enabling high-fidelity reconstruction of individual samples even under large batch sizes. Nevertheless, existing approaches often require architectural modifications, which limit their practical applicability. In this work, we bridge this gap by introducing the Activation REcovery via Sparse inversion (ARES) attack, an active GIA designed to reconstruct training samples from large training batches without requiring architectural modifications. Specifically, we formulate the recovery problem as a noisy sparse recovery task and solve it using the generalized Least Absolute Shrinkage and Selection Operator (Lasso). To extend the attack to multi-sample recovery, ARES incorporates the imprint method to disentangle activations, enabling scalable per-sample reconstruction. We further establish the expected recovery rate and derive an upper bound on the reconstruction error, providing theoretical guarantees for the ARES attack. Extensive experiments on CNNs and MLPs demonstrate that ARES achieves high-fidelity reconstruction across diverse datasets, significantly outperforming prior GIAs under large batch sizes and realistic FL settings. Our results highlight that intermediate activations pose a serious and underestimated privacy risk in FL, underscoring the urgent need for stronger defenses.

Recent advancements in diffusion-based image editing pose a significant threat to the authenticity of digital visual content. Traditional embedding-based watermarking methods often introduce perceptible perturbations to maintain robustness, inevitably compromising visual fidelity. Meanwhile, existing zero-watermarking approaches, typically relying on global image features, struggle to withstand sophisticated manipulations. In this work, we uncover a key observation: while individual image patches undergo substantial alterations during AI-based editing, the relational distance between patch pairs remains relatively invariant. Leveraging this property, we propose Relational Zero-Watermarking (Rel-Zero), a novel framework that requires no modification to the original image but derives a unique zero-watermark from these editing-invariant patch relations. By grounding the watermark in intrinsic structural consistency rather than absolute appearance, Rel-Zero provides a non-invasive yet resilient mechanism for content authentication. Extensive experiments demonstrate that Rel-Zero achieves substantially improved robustness across diverse editing models and manipulations compared to prior zero-watermarking approaches.

Recent advancements in AI-generated content (AIGC) have introduced new challenges in intellectual property protection and the authentication of generated objects. We focus on scenarios in which an author seeks to assert authorship of an object generated using latent diffusion models (LDMs), in the presence of adversaries who attempt to falsely claim authorship of objects they did not create. While proof-of-ownership has been studied in the context of multimedia content through techniques such as time-stamping and watermarking, these approaches face notable limitations. In contrast to traditional content creation sources (e.g., cameras), the LDM generation process offers greater control to the author. Specifically, the random seed used during generation can be deliberately chosen. By binding the seed to the author's identity using cryptographic functions, the author can assert to be the creator of the object. We refer to this stronger guarantee as proof-of-authorship, since only the creator of the object can legitimately claim the object. This contrasts with proof-of-ownership via time-stamping or watermarking, where any entity could potentially claim ownership of an object by being the first to timestamp or embed the watermark. We propose a proof-of-authorship framework involving a probabilistic adjudicator who quantifies the probability that a claim is false. Furthermore, unlike prior approaches, the proposed framework does not involve any secret. We explore various attack scenarios and analyze design choices using Stable Diffusion 2.1 and XL as representative case studies.

Function Secret Sharing (FSS) schemes enable sharing efficiently secret functions. Schemes dedicated to point functions, referred to as Distributed Point Functions (DPFs), are the center of FSS literature thanks to their numerous applications including private information retrieval, anonymous communications, and machine learning. While two-party DPFs benefit from schemes with logarithmic key sizes, multi-party DPFs have seen limited advancements: $O(\sqrt{N})$ key sizes (with $N$, the function domain size) and/or exponential factors in the key size. We propose a DDH-based technique reducing the key size of existing multi-party schemes. In particular, we build an honest-majority DPF with $O(\sqrt[3]{N})$ key size. Our benchmark highlights key sizes up to $10\times$ smaller (on realistic problem sizes) than state-of-the-art schemes. Finally, we extend our technique to schemes supporting comparison functions.

Autonomous AI agents powered by large language models are being deployed in production with capabilities including shell execution, file system access, database queries, and multi-party communication. Recent red teaming research demonstrates that these agents exhibit critical vulnerabilities in realistic settings: unauthorized compliance with non-owner instructions, sensitive information disclosure, identity spoofing, cross-agent propagation of unsafe practices, and indirect prompt injection through external resources [7]. In healthcare environments processing Protected Health Information, every such vulnerability becomes a potential HIPAA violation. This paper presents a security architecture deployed for nine autonomous AI agents in production at a healthcare technology company. We develop a six-domain threat model for agentic AI in healthcare covering credential exposure, execution capability abuse, network egress exfiltration, prompt integrity failures, database access risks, and fleet configuration drift. We implement four-layer defense in depth: (1) kernel level workload isolation using gVisor on Kubernetes, (2) credential proxy sidecars preventing agent containers from accessing raw secrets, (3) network egress policies restricting each agent to allowlisted destinations, and (4) a prompt integrity framework with structured metadata envelopes and untrusted content labeling. We report results from 90 days of deployment including four HIGH severity findings discovered and remediated by an automated security audit agent, progressive fleet hardening across three VM image generations, and defense coverage mapped to all eleven attack patterns from recent literature. All configurations, audit tooling, and the prompt integrity framework are released as open source.

Computer use agents create new privacy risks: training data collected from real websites inevitably contains sensitive information, and cloud-hosted inference exposes user screenshots. Detecting personally identifiable information in web screenshots is critical for privacy-preserving deployment, but no public benchmark exists for this task. We introduce WebPII, a fine-grained synthetic benchmark of 44,865 annotated e-commerce UI images designed with three key properties: extended PII taxonomy including transaction-level identifiers that enable reidentification, anticipatory detection for partially-filled forms where users are actively entering data, and scalable generation through VLM-based UI reproduction. Experiments validate that these design choices improve layout-invariant detection across diverse interfaces and generalization to held-out page types. We train WebRedact to demonstrate practical utility, more than doubling text-extraction baseline accuracy (0.753 vs 0.357 mAP@50) at real-time CPU latency (20ms). We release the dataset and model to support privacy-preserving computer use research.

Federated computing (FC) enables collaborative computation such as machine learning, analytics, or data processing across distributed organizations keeping raw data local. Built on four architectural pillars, distributed data assets, federated services, standardized APIs, and decentralized services, FC supports sovereignty-preserving collaboration. However, federated systems spanning organizational and jurisdictional boundaries lack a portable mechanism for enforcing sovereignty-critical constraints. They often depend on runtime policy evaluation, shared trust infrastructure, or institutional agreements that introduce coordination overhead and provide limited cryptographic assurance. Federated Computing as Code (FCaC) is a declarative architecture that addresses this gap by compiling authority and delegation into cryptographically verifiable artifacts rather than relying on online policy interpretation. Boundary admission becomes a local verification step rather than a policy decision service. FCaC separates constitutional governance from procedural governance. Admission is validated locally at execution boundaries using proof-carrying capabilities, while stateful services may still implement post-admission controls such as ABAC, risk scoring, quotas, and workflow state. FCaC introduces Virtual Federated Platforms (VFPs), which combine Core, Business, and Governance contracts through a cryptographic trust chain: Key Your Organization (KYO), Envelope Capability Tokens (ECTs), and proof of possession (PoP). We demonstrate the approach in a proof-of-concept cross-silo federated learning workflow using MNIST as a surrogate workload to validate the admission mechanisms and release an open-source implementation showing envelope issuance, boundary verification, and envelope-triggered training.

Retrieval-Augmented Generation (RAG) systems introduce a critical vulnerability: contextual leakage, where adversaries exploit instruction-following to exfiltrate Personally Identifiable Information (PII) via adaptive extraction. Current defenses force a rigid trade-off between semantic utility and latency. We present SEAL-Tag, a privacy-preserving runtime environment that resolves this via a Verify-then-Route paradigm. SEAL-Tag introduces the SEAL-Probe protocol, transforming auditing into a structured tool-use operation where the model generates a verifiable PII-Evidence Table (PET) alongside its draft. To adjudicate this evidence, we employ a Probabilistic Circuit (PC) that enforces verifiable logical constraints for robust decision-making. To overcome the privacy "Cold Start" problem, we introduce the S0–S6 Anchored Synthesis Pipeline, generating high-fidelity, provenanced RAG interactions. We pair this with a Two-Stage Curriculum that first optimizes for entity detection before aligning the model to the rigorous audit protocol. Our evaluation demonstrates that SEAL-Tag establishes a new Pareto frontier, reducing adaptive leakage by over 8$\times$ while matching the utility and speed of unsafe baselines.

Cyber deception assists in increasing the attacker's budget in reconnaissance or any early phases of threat intrusions. In the past, numerous methods of cyber deception have been adopted, such as IP address randomization, the creation of honeypots and honeynets mimicking an actual set of services, and networks deployed within an enterprise or operational technology(OT) network. These types of strategies follow naive approaches of recreating services that are expensive and that need a lot of human intervention. The advent of cloud services and other automations of containerized applications, such as Kubernetes, makes cyber defense easier. Yet, there remains a lot of potential to improve the accuracy of these deception strategies and to make them cost-effective using artificial intelligence (AI)-based solutions by making the deception more dynamic. Hence, in this work, we review various AI-based solutions in building network- and device-level cyber deception methods in contested environments. Specifically, we focus on leveraging the fusion of large language models (LLMs) and reinforcement learning(RL) in optimally learning these cyber deception strategies and validating the efficacy of such strategies in some stealthy attacks against OT systems in the literature.

Attacks can exploit zero-day or one-day vulnerabilities that are not publicly disclosed. To detect these vulnerabilities, security researchers monitor development activities in open-source repositories to identify unreported security patches. The sheer volume of commits makes this task infeasible to accomplish manually. Consequently, security patch detectors commonly trained and evaluated on security patches linked from vulnerability reports in the National Vulnerability Database (NVD). In this study, we assess the effectiveness of these detectors when applied in-the-wild. Our results show that models trained on NVD-derived data show substantially decreased performance, with decreases in F1-score of up to 90% when tested on in-the-wild security patches, rendering them impractical for real-world use. An analysis comparing security patches identified in-the-wild and commits linked from NVD reveals that they can be easily distinguished from each other. Security patches associated with NVD have different distribution of commit messages, vulnerability types, and composition of changes. These differences suggest that NVD may be unsuitable as the sole source of data for training models to detect security patches. We find that constructing a dataset that combines security patches from NVD data with a small subset of manually identified security patches can improve model robustness.

Privacy protection mechanisms are a fundamental aspect of security in cryptocurrency systems, particularly in decentralized networks such as Bitcoin. Although Bitcoin addresses are not directly associated with real-world identities, this does not fully guarantee user privacy. Various deanonymization solutions have been proposed, with network layer deanonymization attacks being especially prominent. However, existing approaches often exhibit limitations such as low precision. In this paper, we propose NTSSL, a novel and efficient transaction deanonymization method that integrates network traffic analysis with semi-supervised learning. We use unsupervised learning algorithms to generate pseudo-labels to achieve comparable performance with lower costs. Then, we introduce NTSSL+, a cross-layer collaborative analysis integrating transaction clustering results to further improve accuracy. Experimental results demonstrate a substantial performance improvement, 1.6 times better than the existing approach using machining learning.

Agentic LLM systems equipped with persistent memory, RAG pipelines, and external tool connectors face a class of attacks - Logic-layer Prompt Control Injection (LPCI) - for which no automated red-teaming instrument existed. We present LAAF (Logic-layer Automated Attack Framework), the first automated red-teaming framework to combine an LPCI-specific technique taxonomy with stage-sequential seed escalation - two capabilities absent from existing tools: Garak lacks memory-persistence and cross-session triggering; PyRIT supports multi-turn testing but treats turns independently, without seeding each stage from the prior breakthrough. LAAF provides: (i) a 49-technique taxonomy spanning six attack categories (Encoding~11, Structural~8, Semantic~8, Layered~5, Trigger~12, Exfiltration~5; see Table 1), combinable across 5 variants per technique and 6 lifecycle stages, yielding a theoretical maximum of 2,822,400 unique payloads ($49 \times 5 \times 1{,}920 \times 6$; SHA-256 deduplicated at generation time); and (ii) a Persistent Stage Breaker (PSB) that drives payload mutation stage-by-stage: on each breakthrough, the PSB seeds the next stage with a mutated form of the winning payload, mirroring real adversarial escalation. Evaluation on five production LLM platforms across three independent runs demonstrates that LAAF achieves higher stage-breakthrough efficiency than single-technique random testing, with a mean aggregate breakthrough rate of 84% (range 83–86%) and platform-level rates stable within 17 percentage points across runs. Layered combinations and semantic reframing are the highest-effectiveness technique categories, with layered payloads outperforming encoding on well-defended platforms.

Artificial Intelligence (AI) systems are increasingly dependent on complex, multi-layered software supply chains that introduce challenges for reproducibility, transparency, and security assurance. This study presents an Artificial Intelligence Bill of Materials (AIBOM) schema extending the CycloneDX standard to capture AI-specific provenance, model lineage, and disclosure metadata. The framework provides a formalised approach to verifiable software provenance through structured schema engineering, cryptographic validation, and agent-driven automation. An autonomous AI pipeline is developed to perform continuous environment inspection, vulnerability enrichment, and reproducibility auditing using machine-verifiable provenance chains. Empirical evaluation demonstrates 98.7% reproducibility fidelity, 96.2% vulnerability match precision, and a 63% reduction in manual oversight across containerised analytic workflows. These results confirm the feasibility of automated provenance assurance and reproducible AI lifecycle validation. The AIBOM framework advances the scientific foundations of software supply chain transparency and AI reproducibility engineering, offering a generalisable methodology for securing AI systems, strengthening provenance integrity, and supporting compliance with international information security standards.

Reinforcement learning techniques are being explored as solutions to the threat of cyber attacks on enterprise networks. Recent research in the field of AI in cyber security has investigated the ability of homogeneous multi-agent reinforcement learning agents, capable of inter-agent communication, to respond to cyberattacks. This paper advances the study of learned communication in multi-agent systems by examining heterogeneous agent capabilities within a simulated network environment. To this end, we leverage CommFormer, a publicly available state-of-the-art communication algorithm, to train and evaluate agents within the Cyber Operations Research Gym (CybORG). Our results show that CommFormer agents with heterogeneous capabilities can outperform other algorithms deployed in the CybORG environment, by converging to an optimal policy up to four times faster while improving standard error by up 38%. The agents implemented in this project provide an additional avenue for exploration in the field of AI for cyber security, enabling further research involving realistic networks.

We present NanoZK, a zero-knowledge proof system for verifiable LLM inference: clients and third-party auditors check that a provider executed the advertised model on a committed input without learning weights or activations. NanoZK introduces a layerwise proof framework that decomposes transformer inference into independently provable layers linked by a SHA-256 commitment chain, yielding constant-size sub-circuit proofs (3.5-3.7 KB; about 83 KB total at L=12), comparable in total size to and substantially more parallelizable than prior ZKML's monolithic 101-126 KB proofs. We prove compositional soundness and zero-knowledge under standard assumptions, design 16-bit lookup-table approximations for softmax, GELU, and normalization with measured perplexity degradation below 1e-4 across six model/dataset combinations, and add a Fisher-information-guided audit-budget triage as an efficiency tool (full soundness still requires verifying every layer). On CPU the MLP sub-circuit proves in about 6.3 s prove-only (about 43 s setup plus prove) with about 22 ms verification at any width; attention prove-only time scales from 0.9 s (d=16) to 184 s (d=256); full-block end-to-end proofs are measured to d=128, with a projected GPU time of about 68 s per block at d=768 from measured O(d^2) MSM scaling and a conservative 15-30x GPU-MSM speedup range based on Icicle's published 30x result for n >= 2^20 and extrapolated to the smaller-n regime. Privacy scope: NanoZK hides weights and activations from verifiers and auditors but does not hide the prompt from the prover; this is complementary to HE/MPC.

Retrieval augmented generation systems have become an integral part of everyday life. Whether in internet search engines, email systems, or service chatbots, these systems are based on context retrieval and answer generation with large language models. With their spread, also the security vulnerabilities increase. Attackers become increasingly focused on these systems and various hacking approaches are developed. Manipulating the context documents is a way to persist attacks and make them affect all users. Therefore, detecting compromised, adversarial context documents early is crucial for security. While supervised approaches require a large amount of labeled adversarial contexts, we propose an unsupervised approach, being able to detect also zero day attacks. We conduct a preliminary study to show appropriate indicators for adversarial contexts. For that purpose generator activations, output embeddings, and an entropy-based uncertainty measure turn out as suitable, complementary quantities. With an elementary statistical outlier detection, we propose and compare their detection abilities. Furthermore, we show that the target prompt, which the attacker wants to manipulate, is not required for a successful detection. Moreover, our results indicate that a simple context summary generation might even be superior in finding manipulated contexts.

Code generation large language models (LLMs) are increasingly integrated into modern software development workflows. Recent work has shown that these models are vulnerable to backdoor and poisoning attacks that induce the generation of insecure code, yet effective defenses remain limited. Existing scanning approaches rely on token-level generation consistency to invert attack targets, which is ineffective for source code where identical semantics can appear in diverse syntactic forms. We present CodeScan, the first black-box, vulnerability-specific scanning framework for auditing code generation LLMs, assuming that the defender specifies the target vulnerability classes and provides corresponding task-relevant prompts. CodeScan identifies attack targets by analyzing structural similarities across multiple generations conditioned on different clean prompts. It combines iterative divergence analysis with abstract syntax tree (AST)-based normalization to abstract away surface-level variation and unify semantically equivalent code, isolating structures that recur consistently across generations. CodeScan then applies LLM-based vulnerability analysis to determine whether the extracted structures contain security vulnerabilities and flags the model as compromised when such a structure is found. We evaluate CodeScan against four representative attacks under both backdoor and poisoning settings across three real-world vulnerability classes. Experiments on 117 models spanning three architectures and multiple model sizes demonstrate 97%+ detection accuracy with substantially lower false positives than prior methods.