Papers

This paper presents EduSOC, a lightweight web-based Security Operations Center (SOC) simulator designed for instructor-led cybersecurity education. SOC analysts must triage large volumes of alerts, separate genuine threats from false positives, and communicate decisions under time pressure. Recreating this environment in the classroom is difficult and often impractical for institutions without access to cyber ranges or enterprise security infrastructure. LITE-SOC was developed to provide a simpler alternative. The platform generates continuous streams of synthetic SOC events and offers separate student and instructor views with visualization tools, event annotation, and region-based chat. Instructors control the pacing of the exercise and can inject targeted incidents to guide the scenario. The goal is to give students a practical introduction to SOC workflows such as triage, prioritization, and decision-making without requiring a full operational SOC environment. The platform is intended for use in guided classroom exercises where students collaboratively investigate alerts and practice real-time triage and communication.

Electrocardiogram (ECG)-based biometric recognition has emerged as a promising solution for secure authentication and liveness detection. However, most existing methods rely on unimodal deep learning architectures that independently process either one-dimensional (1D) temporal signals or two-dimensional (2D) time-frequency representations, limiting robustness and generalization. To address this issue, this paper proposes a hybrid framework integrating 1D and 2D convolutional neural networks (CNNs) within a unified end-to-end architecture. The 1D branch extracts temporal and morphological features from raw ECG signals, while the 2D branch captures discriminative spectral information from time-frequency representations. An attention-guided fusion mechanism dynamically weights both modalities according to input characteristics, overcoming the limitations of conventional static fusion strategies. The framework was evaluated on three benchmark datasets (ECG-ID, MIT-BIH, and PTB), including healthy subjects and patients with cardiac pathologies, achieving identification accuracies of 99.56%, 100.00%, and 99.89%, respectively. To assess long-term biometric permanence, experiments were also conducted on the multi-session Heartprint dataset spanning ten years. The proposed approach achieved same-session accuracies of 98.54% (S1), 99.09% (S2), 94.93% (S3R), and 96.08% (S3L), while cross-session evaluations reached 56.33% (S1-S2) and 53.27% (S2-S3R), demonstrating the ability to capture stable biometric signatures over time. The optimal configuration combines InceptionTime for 1D processing, ResNet-34 for 2D analysis, and attention-based fusion. Ablation studies confirm that the proposed attention mechanism consistently outperforms conventional fusion approaches. Overall, the proposed framework provides a robust, scalable, and high-performance solution for ECG biometric recognition.

Prompt injection is the most critical vulnerability in deployed AI agents. Despite recent progress, we show that the prevailing defense paradigm (data-instruction separation) both fails to detect attacks that operate through contextual manipulation and degrades contextually appropriate behavior. We then recast prompt injection via the lens of Contextual Integrity (CI), a privacy theory that judges information flow compliance with contextual norms. This explains types of attacks that current defenses attempt to patch and predict advanced ones future agents will face. We develop unique benign and attack scenarios that force an agent to violate the norms by (1) misrepresenting the flow, (2) manipulating norms, or (3) mixing multiple flows. This reframing suggests an impossibility result: an adversary can always construct a context under which a blocked flow appears legitimate, or a defender who tightens norms will block genuinely legitimate flows. Our findings suggest that current research addresses a shrinking fraction of future attack surfaces. Instead, through CI, we offer a principled framework for evaluating context-sensitive failures, and designing CI-aware alignment for the frontier autonomous agents.

Synthetic facial videos have proliferated across social media faster than platform moderation can respond, raising the cost of disinformation and identity-based attacks. Frame-level deepfake detectors degrade sharply as generator quality increases; high-quality 128x128 GAN output cuts spatial-only accuracy by five percentage points while leaving temporal inconsistencies largely intact. We address this gap with a 3D Convolutional Neural Network detector based on R3D-18, trained with a composite loss that combines binary cross-entropy with a temporal-consistency regularizer. The model processes 16-frame clips from the DeepfakeTIMIT dataset and is initialized from Kinetics-400 action-recognition weights. We report 92.8% accuracy on intra-dataset evaluation at 128x128 resolution; cross-dataset transfer to FaceForensics++ without fine-tuning reaches 76.4%, rising after minimal fine-tuning. Ablation studies show that transfer learning contributes 7.2 percentage points and face tracking adds 3.5 points, while temporal consistency regularization provides additional gains on high-quality fakes. The results establish that temporal artifacts generalize more broadly than spatial ones, providing a detection signal that survives social-media re-encoding.

Network intrusion detection systems play a vital role in protecting networks by detecting malicious network traffic which can then be investigated by a cybersecurity operations centre. State-of-the-art approaches utilise supervised machine learning methods to train a classification model to recognise known cyberattacks; however, these models require a large labelled dataset to train and show poor performance when trained on smaller datasets. In an attempt to address this shortcoming, anomaly detection models learn the distribution of benign traffic and flag non-conforming traffic as malicious. While these methods do not require malicious examples to train, they suffer from high false-positive rates rendering them impractical. As a result, networks may be particularly vulnerable when there are insufficient labelled instances of a specific attack class to train an effective classifier. This often occurs in newly established networks or when previously unseen types of attacks emerge. To address this challenge, this work proposes the use of a triplet network, utilising online triplet mining and a KNN classifier, which is able to perform few-shot classification, enabling effective intrusion detection after being trained on a limited number of malicious examples. Various online triplet mining algorithms were explored and model design choices, such as the inference algorithm and optimised distance metrics, were compared and evaluated through a series of ablation studies. The final model was compared against other state-of-the-art approaches in few-shot binary and multiclass classification, where the proposed approach was found to be competitive with existing methods when trained on as little as 10 malicious samples of each class.

Polynomial multiplication is fundamental in lattice-based cryptography. While the Number Theoretic Transform (NTT) enables fast multiplication, it imposes constraints on the modulus of the coefficient field. Hafiz et al. (2025) addressed this limitation by analyzing the incomplete NTT, which combines a truncated NTT with conventional multiplication methods In this work, we revisit Toom-4 multiplication in the context of incomplete NTT. Although Toom-4 is asymptotically faster than Karatsuba, its precise cost has not been expressed in a form compatible with the incomplete NTT framework. We present a concrete Toom-4 implementation and derive explicit operation counts that separate additions/subtractions and multiplications over the coefficient field. Our analysis based on addition chains yields a simple cost model for incomplete NTT. Using this model, we analyze hybrid strategies combining Toom-4, Karatsuba, and incomplete NTT. We identify parameter ranges where Toom-4 is advantageous and validate the predicted behavior experimentally.

Tool-using LLM agents increasingly rely on external tools to make consequential decisions, yet most existing agent-security benchmarks and defenses implicitly assume that tool feedback is trustworthy once a tool has been selected. We study a different failure mode, cognitive poisoning, in which a malicious tool behaves plausibly during exploration, accumulates trust through benign-looking feedback, and becomes harmful only when hidden state conditions align with the final executable action. To study this setting, we construct TRUST-Bench, a task-conditioned benchmark of 1,970 hidden-trigger tool-compromise episodes with matched safe controls, introduce an asymmetric penalty metric, GuardedJoint, to better reflect real deployment risk, and present VISTA-Guard, a backbone-agnostic framework for final-action risk scoring. The core idea is to abstract multi-step tool interaction into structured environment variables that encode trust-formation dynamics and then score the risk of the final executable action from this trajectory-conditioned representation. Experiments show that prompt-centric heuristics, scalarized features, and zero-shot judges fail in this regime, whereas trajectory-aware final-action scoring yields strong in-domain discrimination and remains effective under balanced out-of-distribution transfer. Under GuardedJoint, VISTA-Guard reaches $84.2$ in-domain and $56.9$ on balanced out-of-distribution evaluation, while methods that optimize only one side of the safety–utility tradeoff collapse to zero. These findings support a broader view of agent security in black-box tool ecosystems: the decisive defense target is not local prompt text or tool descriptors alone, but the way trust is formed across the interaction trajectory and committed through the final action.

As software systems grow increasingly complex, automated vulnerability repair (AVR) remains difficult because the materials available to a repair system are usually failure artifacts rather than repair guidance. Traditional analysis techniques can provide suspicious locations, reduced triggers, or constraints, but they are costly to configure across repositories and seldom directly actionable for patch generation. Recent LLM-based agents can edit and validate repository-level patches, and experience-based systems can reuse prior repair traces or demonstrations, but they still need current-instance evidence that turns a broad, symptom-level failure report into a concrete repair decision. We present ContraFix, an agentic AVR framework that constructs such evidence through contrastive runtime analysis. Starting from a failing witness, ContraFix generates nearby failing and non-failing variants, executes them through aligned probe sites, and compares their runtime states to infer the repair boundary and guide source-level patching. Each candidate patch is accepted only after build and validation. ContraFix also stores validated repair episodes in a dual-track skill base, reusing mutation skills to construct useful variants and correction skills to refine failed patches. On SEC-Bench, ContraFix with GPT-5-mini achieves resolution rate of 92.0% over three repeated runs and an average resolution rate of 91.8% +/- 0.8. On PatchEval, it resolves 73.8% of 225 Go, Python, and JavaScript instances. A semantic audit of benchmark-validated SEC-Bench patches shows that 58.2% of ContraFix's patches are semantically correct, compared with 31.3% for the strongest baseline, indicating that the proposed framework improves semantic correctness beyond benchmark validation.

Large language models (LLMs) are commonly adapted to downstream tasks through fine-tuning, but fine-tuning data often contains sensitive information that may be leaked by the resulting model. Differential privacy (DP) offers formal protection against such leakage, yet DP fine-tuning of LLMs still suffers from substantial utility degradation due to gradient clipping and noise injection. Existing work improves this trade-off by combining DP with parameter-efficient fine-tuning methods such as LoRA, which constrain the form of updates. In this work, we study a complementary direction: selective fine-tuning, which constrains where updates are applied. We propose DP-SelFT, a framework for differentially private selective fine-tuning of LLMs. DP-SelFT addresses three DP-specific challenges in parameter selection: avoiding repeated privacy cost, improving stability under noisy estimates, and selecting parameters that remain useful under clipped and noisy updates. It first constructs a lightweight DP synthetic dataset and performs selection only on this synthetic data, so the selection stage incurs no additional privacy cost. It then conducts layer-level selection by temporarily training candidate layer subsets on a synthetic training split and evaluating them on a synthetic validation split. Crucially, this temporary training is performed under a perturbation regime matched to downstream DP fine-tuning, with worst-case perturbations of the same scale as DP noise. This favors layer subsets that are not only learnable but also robust to noisy private updates. Experiments on benchmark tasks show that DP-SelFT consistently improves the privacy–utility trade-off over existing DP fine-tuning baselines under the same privacy guarantees.

Safety-aligned language models often refuse cybersecurity requests whose wording resembles misuse, even when the task is authorized and defensive. This makes security evaluation ambiguous: a failed answer may reflect missing capability or refusal-policy intervention. Ablating Safety studies alignment removal as a controlled transformation-evaluation protocol for authorized security tasks, comparing authorized-context prompting, reversible refusal-direction activation projection, representation-control projections, and LoRA-based de-alignment or task adaptation. We evaluate refusal, attempt rate, validated security success, general-capability retention, instability, and out-of-scope unsafe compliance on Security-AR, a 60-prompt suite of authorized security, benign general, and non-operational spillover probes. The reported runs include a four-model projection pilot with 416 completions, a three-model Qwen2.5 LoRA extension with 1,980 held-out completions, representation and robustness sweeps, and executable secure-repair validators. Single-vector refusal projection raises mean security score only from 0.46 to 0.50 while increasing unsafe compliance from 0.10 to 0.47; rank-4 refusal-subspace projection reaches 0.51 while matching the aligned spillover rate. Task-only LoRA raises mean security score to 0.87 with general score 0.83 and unsafe compliance 0.13, while refusal-suppression with retention raises spillover to 0.27. These results support evaluating alignment removal as a utility-risk frontier, not as an uncensoring recipe, and treating compliance alone as neither competence nor safe deployment.

We present a quantum attack on ML-KEM and related 2-power cyclotomic lattice schemes. Combining with Parts I-III, we provide an algorithm and verify the resulting approximation factor satisfies $\gamma\le 21 < q/2=1664.5$ for ML-KEM-1024, with a success probability $\ge 0.99$. We apply a tower decomposition of the Principal Ideal Problem (PIP) through the chain $\Q\subset \Q(\zeta_8)\subset\cdots\subset \Q(\zeta_{2^k})$ which yields a polynomial-time quantum algorithm costing $O(n^3 \log^2 n)$ gates, $O(n^2 \log n)$ qubits, and poly$(n)$ classical bit operations. We extend the analysis to Falcon, Hawk, and NTRU over 2-power cyclotomic rings with polynomial-time quantum algorithms.

Side-channel attacks exploit unintended information leakage from system behavior and continue to pose serious privacy risks in modern platforms. Despite extensive prior work, side-channel analysis remains largely manual and fragmented, typically assuming predefined target events and a fixed set of known channels. As systems and applications grow increasingly complex, several fundamental questions remain unanswered: which user or system events are sensitive in practice, how side channels associated with these events can be systematically discovered without exhaustive manual effort, and how their leakage can be analyzed at scale without prohibitive data collection and model training costs. To address these questions, we present SCAgent, an automated framework for side-channel risk analysis. To identify sensitive targets beyond manually specified events, SCAgent performs agent-driven system exploration guided by LLM-based semantic reasoning. To systematically discover side channels while mitigating the risk of LLM hallucination, it reasons over system documentation and incorporates explicit verification to enforce semantic consistency, threat-model feasibility, and per-channel usability. To enable scalable analysis under limited data, SCAgent adopts a few-shot learning paradigm based on foundation models, avoiding the need to train bespoke models for each channel–event pair. To bridge the gap between raw time-series side-channel signals and tabular foundation models, SCAgent further introduces a time-shift–robust feature extraction layer that enables effective downstream analysis. We instantiate SCAgent on iOS as a first step, focusing on OS-level side channels observable by unprivileged applications. Our evaluation spans standard benchmarks such as foreground app and website fingerprinting, as well as newly identified sensitive in-app activities in popular applications.

We prove that the $L^2$ CVP distance from a random short ring element to the log-unit lattice of $\Q(\zeta_{2^k})$ converges to $\frac{\pi}{2\sqrt{6}}\sqrt{n}$ as $n=2^{k-1}\to\infty$. We then show that this target lies inside the Voronoi cell of the origin for $k\ge 4$. For the $L^\infty$ norm, the maximum over $n$ sub-Gaussian coordinates yields $O(\sqrt{\log n})$ which translates into a sub-polynomial approximation factor for the Short Generator Problem. We show a Coarse Lattice Theorem that Babai's algorithm returns zero for all structured targets, yet exactly recovers unit perturbations of arbitrary size. For module determinant ideals, we further prove the Trigamma Theorem that proves an intrinsic imbalance $\sigma_{g_0}=O(1)$ independent of the modulus $q$. Finally, combined with Parts I and II, we reduce the CDPR factor for ML-KEM from $\exp(\tO(\sqrt{n}))$ to a sub-polynomial value.

We present the Agentic AI Detection and Response (ADR) system, the first large-scale, production-proven enterprise framework for securing AI agents operating through the Model Context Protocol (MCP). We identify three persistent challenges in this domain: (1) limited observability – existing Endpoint Detection and Response (EDR) tools see file writes but not the agent reasoning, prompts, or causal chains linking intent to execution; (2) insufficient robustness – static defenses constrained by pre-defined rules fail to generalize across diverse attack techniques and enterprise contexts; and (3) high detection costs – LLM-based inference is prohibitively expensive at scale. ADR addresses these challenges via three components: the ADR Sensor for high-fidelity agentic telemetry, the ADR Explorer for systematic pre-deployment red teaming and hard-example generation, and the ADR Detector for scalable, two-tier online detection combining fast triage with context-aware reasoning. Deployed at Uber for over ten months, ADR has sustained reliable detection in production with growing adoption reaching over 7,200 unique hosts and processing over 10,000 agent sessions daily, uncovering hundreds of credential exposures across 26 categories and enabling a shift-left prevention layer (97.2% precision, 206 detected credentials). To validate the approach and enable community adoption, we introduce ADR-Bench (302 tasks, 17 techniques, 133 MCP servers), where ADR achieves zero false positives while detecting 67% of attacks – outperforming three state-of-the-art baselines (ALRPHFS, GuardAgent, LlamaFirewall) by 2–4x in F1-score. On AgentDojo (public prompt injection benchmark), ADR detects all attacks with only three false alarms out of 93 tasks.

DRAM scaling has exacerbated the RowHammer vulnerability. To counter this, JEDEC recently introduced Per Row Activation Counting (PRAC) with the Alert Back-Off protocol as an optional DDR5 feature. While promising, PRAC requires per-row counter cells that incur area overhead, and updating them on every activation lengthens DRAM timing parameters, degrading performance. Probabilistic mitigations such as MINT offer a lower-cost alternative by randomly selecting and mitigating rows within periodic mitigation windows. MINT is effective at higher thresholds (>= 1000), but at lower thresholds, it must raise its mitigation rate to overcome the non-selection problem, where heavily hammered rows can repeatedly escape sampling. This fixed-rate scaling reduces effective memory bandwidth even when no attack is present. To overcome this limitation, we propose PrISM, an intersection-based probabilistic mitigation that correlates sampled rows across windows using a Sampled History Queue (SHQ). PrISM samples a few activation slots per window, stores sampled-but-unmitigated rows in the SHQ, and requests an additional mitigation through the existing Alert Back-Off protocol when a sampled row reappears in this history. This allows PrISM to increase mitigation only when persistent row activity is observed, without globally increasing the fixed mitigation rate. At the threshold of 500, PrISM incurs a negligible 0.2% average slowdown compared to 14% for PRAC, with no DRAM array changes or per-row counters and only 625B of SRAM per bank, one to two orders of magnitude less than prior secure counter-based in-DRAM defenses. Compared to MINT, PrISM provides better scalability at low thresholds, reducing average slowdown from 10.7% to 1.5% at a threshold of 250, a 7.1x reduction. PrISM is open-sourced at https://github.com/STAR-Laboratory/prism.

Guardrails are a critical safety layer for modern AI systems, but their operating regime is changing. As LLMs are deployed as customized assistants, safety policies are increasingly specified at inference time by users, organizations, or regulatory contexts. This makes safety enforcement fundamentally dynamic: the guardrail should adapt to changing safety policies without retraining. Yet this requirement creates a fundamental tension: faithfully judging complex policy contexts demands reasoning capability, while practical deployment requires low-latency responses. We introduce Latent Policy Guardrail (LPG), a guardrail framework that learnssemantic latent deliberation over dynamic policies. LPG compresses the internal deliberation needed for intent interpretation and policy grounding into continuous states supervised by decision-relevant semantics. At inference time, it generates only a compact verdict anchored to the violated policy clauses, preserving auditability while avoiding the latency of explicit reasoning. Across policy guardrail benchmarks, LPG-4B reaches 84.5% average safety accuracy and 77.9% F1 by compressing deliberation into just 10 latent tokens, outperforming the strongest dynamic baseline while running roughly 11 times faster than Qwen3-4B-Thinking under the single-sample evaluation setup. Code and data are available at https://github.com/SaFo-Lab/Latent_Policy_Guard.

Critical infrastructure defense is fundamentally bottlenecked by the operational reality that preventive controls are frequently bypassed by sophisticated supply-chain compromises and stolen administrative credentials. When prevention fails, defense relies entirely on rapid, post-ingress threat detection and automated response across sovereign sectors. We present a novel, federated, high-throughput stream-processing and correlation framework designed to detect coordinated cross-sector threat campaigns and orchestrate containment at machine speed. By utilizing a stateless Pre-Filtering Dispatcher Subsystem (PFDS), in-memory lock-sharded state workers, and a 95% statistical watermark heuristic, our system maintains detection momentum during network partitions to evacuate speculative alerts. Delayed telemetry is subsequently reconciled directly within a version-keyed columnar storage engine via deterministic time-bucket hashing, eliminating state-retraction overhead. We evaluate a prototype of our framework - implemented in Go with an instantiated production-grade columnar analytical store - against a 500,000 events per second workload. The results demonstrate an internal framework processing overhead of under 7 seconds, while achieving total end-to-end operational convergence - accounting for multi-sector detection, correlation, wide-area network (WAN) propagation, windowing stability, VLAN-level response, and hardware level mitigation commitment - within a realistic 12-20 seconds window.

Clarification-seeking behavior is widely regarded as a desirable property of LLM agents, enabling them to resolve ambiguity before acting on underspecified tasks. However, the security implications of this interaction pattern remain unexplored. We investigate whether the transition from standard execution to a clarification-seeking state increases an agent's susceptibility to prompt injection attacks. We introduce ASPI (Ambiguous-State Prompt Injection), a benchmark of 728 task-attack scenarios that isolates clarification as a distinct agent state and measures how this state transition affects vulnerability under controlled conditions. Each benchmark instance is evaluated under matched execution and clarification settings: in the execution setting, the agent acts on a fully specified instruction and encounters adversarial content only through tool-returned data; in the clarification setting, the agent must first request and incorporate additional user input before acting. We evaluate ten frontier LLMs and find that clarification-seeking consistently and substantially amplifies vulnerability. For instance, attack success rises from 1.8% to 34.0% for o3 and from 2.2% to 35.7% for Gemini-3-Flash. A decomposition analysis reveals that this gap reflects both a state-dependent shift in how models process incoming content and a channel-specific effect arising from the agent-solicited clarification interface. These findings demonstrate that standard execution-time security evaluation systematically underestimates the attack surface of interactive agents, and that robustness under fully specified tasks does not translate to robustness under ambiguity. For reproducibility, our data and source code are available at https://github.com/scaleapi/aspi.

Large Language Model (LLM) cascade systems are designed to balance efficiency and performance by processing queries with lightweight models while selectively escalating complex cases to more powerful ones. Such systems seek to reduces computational cost and latency while maintaining task performance, making it an appealing choice for large-scale deployment. However, the cascade design introduces new vulnerabilities through an expanded attack surface: the inclusion of lightweight front-end models and internal decision mechanisms introduces new weaknesses. In this work, we present the first study demonstrating that LLM cascade systems are susceptible to targeted adversarial manipulation, which disrupts both performance objectives and the intended cost advantages of the cascade design. We propose a novel attack framework that employs constrained sequential collaborative optimization of adversarial suffix under cascade dependencies, enabling simultaneous exploitation of lightweight models and decision mechanisms. This framework adapts to adversaries with varying capabilities, inducing controllable degradation in both cost-efficiency and accuracy. Unlike prior attacks targeting standalone models, our approach strategically leverages the cascade structure to achieve significantly stronger impact. Extensive experiments across diverse datasets and representative LLM cascade systems validate the practicality and severity of this attack. Our findings highlight the urgent need to rigorously scrutinize the security of LLM cascade systems and call for broader attention to the systemic risks inherent in such designs.

Computations can be directly carried out over ciphertexts using homomorphic encryption (HE), which is indispensable for privacy-preserving cloud computing. Linear transformation is widely used in neural networks, including large language models. However, the implementation of linear transformation over HE requires a large number of ciphertext rotations, which incur significant memory and hardware overhead despite existing simplification techniques. This paper proposes a triple-hoisted baby-step giant-step algorithm that decomposes the baby step further to substantially reduce the number of ciphertext rotations needed for the CKKS HE evaluation of linear transformation. Moreover, to reduce off-chip memory access, which contributes to the majority of the latency, a memory-optimized data path is proposed by partitioning the algorithm into multiple phases. Furthermore, an efficient FPGA-based hardware accelerator with an optimized permutation circuit for message routing is designed for the proposed scheme. For a set of typical parameters, the proposed design reduces the off-chip memory access by 2.9x compared to the best prior design. Synthesized for Xilinx Virtex UltraScale+ devices, the proposed design achieves a 5.8x reduction in computational latency compared with the baseline design.

Artificial Intelligence (AI) is widely adopted today for its ability to detect patterns, automate tasks, and reduce time and cost across various applications. Its integration into Cybersecurity has garnered significant attention, particularly in areas such as intrusion detection, malware analysis, and phishing or spam detection. As AI and cybersecurity evolve, new methods and approaches emerge regularly. Current trends include the use of Generative AI, Natural Language Processing, Federated Learning for privacy-preserving collaborative training, and eXplainable AI to ensure interpretability and trust, which are vital in cybersecurity. This paper presents an interesting review of current AI-based cybersecurity trends, focusing on intrusion detection approaches and aiming to uncover meaningful insights through comparative analysis based on the employed AI techniques and reported performance.

Social engineering attacks exploit human trust rather than software vulnerabilities, making them difficult to detect using conventional filters. We propose a two-stage filter-then-verify framework combining inductive Graph Neural Networks (GNNs) for structural anomaly detection with a co-attention ModernBERT model for content verification. The GNN identifies anomalous sender-receiver patterns, while BERT analyzes message context to reduce false positives. Using the Enron dataset augmented with realistic synthetic campaigns, we show that the framework achieves 86% recall in structural filtering and over 92% precision after BERT refinement, effectively detecting both external attacks and insider threats. Our results demonstrate that combining structural and content analysis allows practical, scalable detection of multi-stage social engineering attacks in email networks.

Traditional cybersecurity methodologies target deterministic systems and fail to address the probabilistic nature of AI, leaving systems vulnerable to attack vectors such as model inversion, data poisoning, and prompt injection. Recent industry reports indicate that a majority of organizations deploying AI lack a dedicated security strategy, with adversarial attacks increasing rapidly year-over-year. We present STRIDE-AI, a framework that bridges the gap between high-level risk standards (NIST AI RMF) and technical vulnerability taxonomies (OWASP LLM Top 10). The framework defines a six-phase assessment lifecycle, introduces a threat modeling adaptation of classical STRIDE for AI systems, and is operationalized through a purpose-built web tool. We provide an initial validation of the approach through a black-box assessment of a deployed LLM chatbot, which successfully reduced the attack success rate from 80% to 15% in our sandbox case study.

Jailbreak attacks on large models have drawn growing attention due to their close ties to societal safety. This work identifies a practical yet unexplored jailbreak scenario, the wide-net-casting scenario, where an adversary can query a group of large models instead of a single one to elicit harmful outputs. Our analysis reveals substantial yet previously overlooked safety risks under this scenario. As a key part of our analysis, we further develop a novel jailbreak method tailored to the wide-net-casting scenario. With this tailored method, the jailbreak success rate can even reach 100% in some experiments when targeting the large models without additional safeguards, exposing wide-net-casting as a distinct, high-risk scenario that warrants attention in future evaluation and defense research.

Post-Quantum cryptography is about to substitute current cryptographic schemes as being resilient in attacks from quantum computers. McEleiece and Bit Flip Key Encapsulation (BIKE) are two delight representatives based on coding theory where classical structural attacks against these algorithms can be successfully phased out by selecting the appropriate key size. Using low cost equipment, the method of Simple Power Analysis (SPA) is used in this paper to evaluate whether or not there is significant information leakage during the decapsulation phase where the shared secret key is generated. Executing a related experiment it is shown that correlation between electromagnetic emissions and secret values exists. In the aftermath, with only 200 power traces collected, machine learning models can predict secret bits of the shared session key, produced during the decapsulation.

AI-enabled Security Orchestration, Automation, and Response (SOAR) systems increasingly employ autonomous agents for cyber defense, yet their resilience to adaptive adversaries is underexplored. We introduce an autonomous red teaming framework that integrates large language models (LLMs) with reinforcement learning (RL) to generate adaptive, multi-stage attack campaigns against autonomous defenders in enterprise networks. A hierarchical design combines an LLM-based planner for strategic intent with an RL controller for tactical execution, supported by reward shaping aligned with kill-chain progression. Evaluation in a high-fidelity enterprise simulation demonstrates the effectiveness of the proposed approach, while also showing that standalone LLM agents fail to sustain multi-stage attack campaigns and that domain-specific cybersecurity models achieve only limited levels of compromise, highlighting the necessity for hybrid LLM-RL approaches to red teaming.

Spracklen et al. (USENIX Security '25) showed that code-generating large language models hallucinate package names that do not exist on PyPI or npm at rates ranging from 5.2% on commercial models to 21.7% on open-source models, creating an attack surface for slopsquatting – the registration of malicious packages under hallucinated names. We replicate their methodology on five frontier code-capable LLMs released between October 2025 and March 2026: Claude Sonnet 4.6, Claude Haiku 4.5, GPT-5.4-mini, Gemini 2.5 Pro, and DeepSeek V3.2. Across 199,845 paired Python and JavaScript prompts validated against PyPI and npm master lists, we measure overall hallucination rates between 4.62% (Claude Haiku 4.5) and 6.10% (GPT-5.4-mini) – an order-of-magnitude compression of the inter-model spread observed by Spracklen, but not a retirement of the threat. Beyond replication, we identify a set of 127 package names (109 on PyPI, 18 on npm) that all five evaluated models invent identically; following coordinated disclosure with PyPI Security and Socket.dev, 53 of these (41 on PyPI, 12 on npm) remain registrable by an attacker after each registry's existing defenses, constituting a model-agnostic supply-chain attack surface that no single-model study can reveal. We further document a Python-over-JavaScript hallucination asymmetry that inverts Spracklen's 2024 finding, identify a Haiku-below-Sonnet inversion within the Anthropic family, and observe a Jaccard-similarity peak between DeepSeek V3.2 and GPT-5.4-mini (J = 0.343) suggestive of shared training-data origins.

FIPS 203, 204 and 205 closed the algorithmic gap in post-quantum cryptography (PQC); the production gap – hybrid combiners, conformance evidence, migration tooling, stateful signing, protocol helpers – remains open. A methodology that scores it must not choose its own dimensions, so we anchor nine to external requirements (CNSA 2.0, CMVP, the TNO CADI survey, the IETF hybrid draft). We present quantum-safe, a hybrid-by-default Python library built against this rubric, and an audit of nine PQC libraries (September 2026). The exercise can lower our own score, and it does. Hybrid combiners exist in three of six verifiable libraries, not the one a March-2026 snapshot found. Discovery and inventory tooling is the sharper gap: none of seven audited libraries has it; quantum-safe adds a CycloneDX CBOM. Only Bouncy Castle holds a CMVP FIPS 140-3 certificate; quantum-safe reports 225/225 runnable ACVP known-answer cases, which is conformance evidence, not validation. Bouncy Castle also supports LMS and XMSS where quantum-safe has LMS only, and quantum-safe's defaults sit below the CNSA 2.0 parameter sets; we score both against ourselves. Every benchmark figure is computed from archived raw runs (bootstrap intervals over 15 runs). A full X25519 + ML-KEM-768 handshake takes a median of 231 microseconds under Docker/Linux (95% interval 218-250), 5.3x an X25519-only handshake and 0.46-2.3% of a typical TLS 1.3 budget. Throughput is flat from 100 to 5,000 threads but below the single-thread rate: threads do not speed up the hybrid path, although liboqs does release the GIL in longer calls (ML-DSA-65 signing scales 2.6x on four threads). Timing side channels are treated in a companion paper.

Standard PII filters often miss contextual data leakage in RAG systems, such as non-regulated attribute clusters that collectively identify individuals. We introduce a Privacy Policy Enforcement (PPE) framework using dual one-class density estimators with fused text embeddings and a calibrated abstain region for out-of-distribution inputs. Using an axis-stratified, multi-LLM synthetic data pipeline across medicine, finance, and law, we found that traditional Gaussian Mixture baselines fail on borderline-safe stress tests by focusing on linguistic register rather than content. Our proposed T3+OCSVM detector, trained on safe and borderline-safe data, achieves a borderline AUROC of 0.93+ while reducing false positives by 44-55 percentage points and maintaining millisecond latency. Compared to supervised MLP classifiers or 14B-parameter LLM judges, our framework offers superior operational suitability, as the former suffers from high abstention rates and the latter from latency and calibration issues. This methodology provides a robust stress-testing standard for any synthetic-data-trained classifier.

This position paper argues that securing LLM agents requires first defining an end-to-end correctness property that specifies when an agent's execution faithfully reflects the user's intent. Modern LLM agents operate over an intent-to-execution pipeline, where natural-language instructions are translated into concrete system operations such as tool calls, API requests, and code execution. While recent defenses have made progress in constraining how agents construct tool calls, most existing formulations implicitly assume that tools are trusted. The emergence of systems such as OpenClaw, with open ecosystems of third-party skills and direct access to user environments, breaks this assumption and exposes new failure modes, including malicious or over-privileged components in the execution pipeline. Despite rapid progress in defense mechanisms, there is no adequate correctness property that defines what “secure” means for LLM agents, nor a principled way to evaluate the coverage of existing defenses. We observe that LLM agents are structurally analogous to compilers, where security violations correspond to mis-executions that do not preserve user intent. Drawing on this analogy, we identify two fundamental problem sources – untrusted data ingestion and untrusted tool execution – and derive four integrity properties that must hold simultaneously: Tool Integrity, Instruction Integrity, Judgment Integrity, and Data Flow Integrity. We call their conjunction intent-to-execution integrity. Analyzing existing agentic defenses against these properties reveals that current systems provide only partial and non-compositional coverage, leaving fundamental gaps in securing modern LLM agents.