Papers

The widespread integration of Large Language Models (LLMs) necessitates rigorous and systematic safety evaluation. Existing paradigms either rely on constructed benchmarks to assess safety from predefined perspectives, or employ dynamic red-teaming to probe potential vulnerabilities. While effective, these approaches face challenges, as they depend heavily on expert domain knowledge, offer limited systematic guarantees, and are vulnerable to rapid obsolescence. To address these limitations, we introduce a novel framework POLARIS that brings the rigor of specification-based software testing to AI safety. POLARIS first compiles unstructured natural-language policies into First-Order Logic (FOL) representations, establishing a traceable link between high-level rules and concrete test cases. This formalization enables the construction of a Semantic Policy Graph, where complex policy violation scenarios are encoded as traversable paths. By systematically exploring this graph, POLARIS uncovers compositional violation patterns, which are then instantiated into executable natural-language test queries, enabling coverage-driven and reproducible safety testing. Experiments demonstrate that POLARIS achieves higher policy coverage and attack success counts compared to established baselines. Crucially, by bridging formal methods and AI safety, POLARIS provides a principled, automated approach to ensuring LLMs adhere to safety-critical policies with verifiable traceability. We release our code at https://github.com/huac-lxy/POLARIS.

Large language model (LLM) safety classifiers such as Llama Guard are effective at detecting overtly harmful prompts but remain vulnerable to adversarial jailbreak attacks that disguise malicious intent through role-play scenarios, fictional framing, and indirect requests. We present Reflect-Guard, a method that augments LLM-based safety classifiers with chain-of-thought self-reflection capabilities through parameter-efficient fine-tuning. Our approach distills analytical reasoning from GPT-4o-mini into structured reflection annotations, then trains Llama-Guard-3-8B via QLoRA to generate logical self-reflections before issuing safety verdicts. Using only 1000 training examples and updating just 0.5% of model parameters (~42M), Reflect-Guard achieves substantial improvements on two challenging benchmarks. On WildGuardTest, F1 score improves from 0.770 to 0.842 (+7.2 pp), with recall on adversarial prompts increasing from 0.513 to 0.921 (+40.8 pp). On JailbreakBench, the attack success rate drops from 10.3% to 1.8%, representing an 82.5% relative reduction. These gains are especially pronounced on adversarial inputs, where the explicit reasoning step enables the model to see through obfuscation techniques that defeat standard pattern-matching approaches. Our results demonstrate that teaching safety classifiers to reason about adversarial intent, rather than simply classify surface patterns, is a promising direction for robust LLM safety.

As Mixture-of-Experts (MoE) architectures are increasingly adopted for scaling Large Language Models (LLMs), safety auditing becomes necessary to verify whether these models produce or facilitate harmful behaviors during operation. However, existing content-based auditing methods typically require access to user prompts, model internals, or outputs, potentially exposing sensitive user information and creating a tension between LLM safety and user privacy. On the other hand, we observe that, in MoE models, different inputs induce different sparse expert-routing patterns, which produce measurable footprints in low-level GPU execution telemetry. We refer to these hardware-observable signals induced by expert-routing decisions as expert routing telemetry; they are derived from GPU execution rather than from router logits or token-level routing assignments. Inspired by this observation, we propose RouteScan, a non-intrusive auditing framework for detecting harmful behaviors through such routing-induced GPU telemetry. Specifically, RouteScan utilizes the number of active GPU threads allocated to expert modules during the prefilling phase as a discriminative micro-architectural fingerprint, and builds a lightweight detection pipeline that isolates cross-domain invariant risk indicators for the precise identification of malicious prompts. Comprehensive evaluations on four open-source MoE LLMs with distinct routing designs demonstrate that RouteScan achieves strong generalization, with an AUROC exceeding 0.91 on unseen harmful domains. Moreover, privacy stress tests show that, although aggregated execution telemetry retains input-related attribute information, full prompts and exact sensitive fields cannot be reliably recovered under the evaluated attacks.

Tabular synthesis is critical for privacy-preserving sharing and augmentation, yet diffusion models rely on implicit mechanisms to capture inter-column relationships. We introduce Geometry-Aware Tabular Diffusion (GATD), which augments tabular diffusion denoisers with pairwise angles and lengths computed from column value differences and used as inputs and auxiliary targets. Our MLP instantiation achieves state-of-the-art benchmark performance while using 3.5x fewer parameters on average (up to 25x for classification tasks): on ten datasets, it wins 8/10 Shape, 7/10 Trend, and 9/10 downstream utility (F1/RMSE), reducing Shape and Trend error by 27% and 20%. Default loss weights transfer to GNN and Transformer denoisers, improving Shape on 27/30 and Trend on 25/30 architecture-dataset cells. A matched ablation shows supervision (not extra inputs or capacity) drives the gain. This shows explicit relational supervision is a portable inductive bias for tabular diffusion.

Frontier AI systems, including large language models and emerging agentic AI tools, offer significant operational benefits but present unique challenges to critical infrastructure (CI) environments due to their non-deterministic and emergent properties. While formal adoption is inherently cautious and tightly controlled due to strict regulatory oversight, widespread accessibility has catalysed shadow AI: the unsanctioned use of frontier AI outside established organisational controls. In CI settings, shadow AI bypasses established assurance and oversight mechanisms, amplifying risks to data protection, decision reliability, and regulatory compliance, with potential consequences for essential service delivery. We present the first empirical study of shadow AI in CI environments, characterising it as a systemic socio-technical condition of assurance erosion. Drawing on semi-structured interviews with senior executives and functional leaders across 27 Australian CI organisations (Communications, Energy, and Water and Sewerage sectors), we analyse how shadow AI manifests in practice, how it interacts with existing technical and governance controls, and the resulting security, assurance, and compliance risks. We develop an empirically derived threat model identifying three primary mechanisms of security degradation: (i) boundary bypass, where data flows circumvent established perimeters; (ii) unassessed capability expansion, where embedded AI features introduce latent risks; and (iii) loss of observability via governance circumvention, undermining forensic auditability and least-privilege enforcement. Our findings demonstrate that shadow AI introduces unmanaged risks that fundamentally challenge existing security and compliance frameworks, necessitating tailored, pathway-aligned governance and control strategies.

Large language models (LLMs) are increasingly applied to cybersecurity question answering (QA) for critical tasks such as incident response and vulnerability analysis. However, real-world operational contexts, including system logs and network configurations, inherently contain sensitive identifiers, e.g., IP addresses, host names, and user accounts. Processing this data with cloud-based models is often unsafe or infeasible in regulated environments. Furthermore, progress in privacy-preserving QA is hindered by the lack of annotated, context-rich datasets capable of jointly evaluating operational reasoning and privacy preservation. To address this gap, we introduce CYBERMASKQA, a privacy-aware QA benchmark covering key security domains. Unlike existing benchmarks that primarily test factual knowledge, CYBERMASKQA grounds questions in realistic organizational contexts with explicit causal dependencies among assets and privileges. Generated through a systematic pipeline, the dataset combines human-curated base scenarios with LLM-driven semantic expansion, annotating each instance with precise private entity labels to enable controlled information disclosure. Evaluations of QA accuracy and masking performance demonstrate the benchmark's utility for developing deployable, context-aware cybersecurity models and facilitating nuanced studies of privacy-utility trade-offs. Upon acceptance, we will release the dataset and the generation framework.

Streaming intrusion-detection studies assemble evaluation streams from network captures by interleaving capture days, pooling captures, or replaying records round robin. We show on two benchmarks that this assembly is an uncontrolled experimental treatment changing what the evaluation measures. On CICIDS2017, reordering an identical record multiset under a fixed positional 70/15/15 split yields held-out samples sharing only 32.5% of their records, at prevalences of 68.235% and 25.2396% (42.9954 points apart), and reverses the measured ordering of the two deterministic scorers. Restricting both arms to the 78000 records both held out removes the reversal, so it is attributable to which records the assembly hands to the test set, not to the order in which the detector saw its history. That attribution assumes that history contributes no more on the records the arms do not share than on those they do. On LITNET-2020, pooling three temporally disjoint captures reports one 6.4982% operating point, the equal-weight mean of per-capture held-out prevalences from 0.176% to 15.7747%, an identity presented as an audit check. The evaluated detector's reset posterior P(r_t=0) equals the hazard rate exactly below the run-length cap, though evaluations spend nearly all their length at or beyond it, and its evaluated score is a function of P(r<=5), not of P(r=0). Its deployed max composition ranks worse than its tail term alone (0.103477 AP, 0.302658 AUC-ROC) because the auxiliary branch is inverted (AUC-ROC 0.281890) and the maximum lets it set the score wherever the tail is small. With evaluated records and fitted model fixed, changing only the accompanying batch moves the ECOD reference implementation's AUC-PR by 0.003063, so published ECOD numbers are not comparable across studies scoring different batches. Every measured value traces to an archived, hash-verified run manifest.

This paper presents CyBOKClaw, an interpretable human-in-the-loop retrieval framework for mapping cybersecurity keywords or phrases (KWoPs) to the Cyber Security Body of Knowledge (CyBOK). Rather than treating the task as strict exact classification, the framework is designed as a top-k candidate generator for expert review. It combines query normalization, curated term expansion, concept-level boosts, topic-description enrichment, and domain-sensitive ranking rules. Because educational KWoPs are often broad, ambiguous, and only approximately aligned with CyBOK terminology, strict exact matching provides only a partial account of practical utility. We therefore evaluate the framework using both structural retrieval metrics and an expert-guided top-5 usefulness metric, ECA-5 (Exact or Closest Acceptable Match at top-5), which records whether the returned candidates contain at least one mapping that an expert would judge exact or accept as the nearest practical CyBOK placement. On the development dataset, CyBOKClaw achieves 64.73% EXA-5 (Exact Match at top-5), 84.18% structural semantic alignment, and 91.88% ECA-5; on the validation dataset, it achieves 81.19% EXA-5, 93.32% structural semantic alignment, and 98.00% ECA-5. These results show that expert-guided top-k usefulness provides a more faithful account of practical CyBOK mapping utility than exact structural matching alone, and that CyBOKClaw is effective as a CyBOK-specific expert-support retrieval system.

Recent demonstrations of large language models producing candidate and confirmed vulnerabilities in production software have renewed the narrative that AI will reshape offensive and defensive security. Headlines emphasize capability; they rarely interrogate costs and incentives. This paper examines LLM-driven vulnerability discovery through a bugonomics lens: the operational economics of producing, proving, prioritizing, and fixing security-relevant defects. Historically, the most visible high-end bugonomics was offense-priced because production-grade zero-days and exploit chains were expensive specialist outputs for governments, brokers, and offensive vendors. Defender-side bugonomics already existed in vulnerability research, reward programs, and vendor remediation work; LLM-assisted systems change its scale and distribution. They make candidate generation, code comprehension, harness construction, proof-of-impact drafting, and report preparation cheaper at codebase scale. Exploits and proofs of concept remain important, but in defender workflows they primarily prove impact, guide prioritization, and justify remediation. The resulting bottleneck is not only finding more bugs; it is absorbing, validating, triaging, patching, and shipping a larger stream of reports. Using public data from Anthropic's Mythos Preview and Mozilla Firefox collaborations, along with public exploit-market price anchors and vulnerability reward programs, we argue that the near-term shift is not simply more zero-days. It is a move toward broader defender remediation throughput: low-signal candidates become cheaper, evidence-rich remediation become more important, and scarce capacity shifts toward maintainer review and release work. The effect is acute in open source, where LLM-assisted discovery can increase report volume while maintainer-side validation, triage, funding, and release capacity may not scale.

Ransomware has grown to become one of the most damaging types of cybercrime, affecting private and public organizations in any sector. While early types of ransomware targeted many victims via automated attacks, ransomware groups have started to specifically target organizations and companies in the expectation of receiving larger ransoms. To increase the pressure on victims, most groups host so-called data leak sites, where information about their victims is made public. The shift towards 'human-operated' ransomware together with easily accessible behavioral traces available from data leak sites makes research investigating operational regularities of ransomware groups of interest. Using leak site posts as behavioral traces of ransomware groups, we created a dataset consisting of over 27,000 posts from 325 groups. Based on this dataset, we analyzed victim concentration, temporal routines and targeting regularities. Our findings suggest that groups do not behave entirely random. Instead, the observable traces found on leak sites show concentration of activity, temporal routines and selective patterns.

Representation engineering (RepE) defenses have shown strong robustness against jailbreak attacks on large language models (LLMs). However, these methods fundamentally rely on black-list supervision: they learn jailbreak-to-refusal activation transformations from harmful or jailbreak data that are inherently incomplete and continuously evolving. Hence, the performance of RepE-based defenses becomes tightly coupled to the quality and coverage of collected harmful samples, leaving models vulnerable to unseen attacks. This reliance also obscures the distinction between defenses that fit known harmful distributions and defenses that protect a benign latent region without estimating the harmful distribution. We adopt the opposite, the white-list perspective, by leveraging the accessibility and abundance of benign data. The goal is to elicit refusal on arbitrary inputs while ensuring that harmless inputs are not falsely rejected. This shifts the core research question to: How can we design a robust benign-latent preservation mechanism such that the benign latent distribution remains intact while refusal is elicited? To answer this, we propose Ellipsoid Control, a test-time defense. It performs projected gradient descent that can elicit refusal on arbitrary inputs, aiming to improve defense effectiveness. At the same time, an anisotropic benign-geometry ellipsoid is fitted from abundant benign data to constrain the update to minimize distortion of the benign latent geometry. This tight constraint helps preserve model utility. Across multiple LLMs, jailbreak attacks, benign tasks, and safety-boundary evaluations, Ellipsoid Control consistently enhances safety while better preserving utility, demonstrating the effectiveness of the white-list approach for jailbreak defense

Cybersecurity awareness training has historically adopted a one-size-fits-all approach, despite established individual differences in how users process and retain security information. Personality has been proposed as one axis along which training content might be tailored; yet no prior study has implemented and empirically evaluated a complete personality-conditional system end-to-end. This paper reports the design, implementation, and quasi-experimental evaluation of TailoredSec, a mobile cybersecurity awareness application that routes training content based on a user's dominant Five-Factor Model (FFM) personality trait, as measured by the ten-item Big Five Inventory (BFI-10). Seventy-four UK-based adults were allocated to a traditional video-training condition ($n = 40$) or a personality-conditional condition ($n = 34$). Both groups completed a four-item scenario-based pre-assessment (scored 0–40), a single training session, and an equivalent post-assessment. The personality-conditional group additionally completed the BFI-10 (Big Five Inventory-10) and was routed to one of four training modules covering five FFM traits (Conscientiousness and Neuroticism share a module). Pre-assessment scores did not differ between groups ($t(69.1) = 0.43$, $p = .67$), confirming baseline equivalence. The personality-conditional group scored significantly higher on the post-assessment ($M = 35.88$, $SD = 5.00$ vs $M = 30.75$, $SD = 10.23$; Welch's $t(58.5) = 2.81$, $p = .007$; Cohen's $d = 0.62$; 95% CI $[1.47, 8.79]$ marks), with a pass-rate of 100% versus 77.5% (Fisher's exact $p < .01$). These results offer preliminary support for personality-conditional content routing as a feasible design principle for cybersecurity awareness training.

This paper examines the erosion of Public Key Cryptography (PKC) security under adaptive adversarial optimisation driven by artificial intelligence. The problem addressed is the growing mismatch between algorithm-centric cryptographic security models and operational attack realities, where adversaries exploit implementation-level observability rather than breaking cryptographic primitives.

Jailbreak prompts can trigger harmful completions on aligned LLMs, In accordance, safety steering has been proposed: test-time activation interventions that steer jailbreak activations to trigger refusal while preserving benign utility. However, existing steering methods are fundamentally supervised and tied to a static, limited training set, whereas real jailbreaks evolve and are often out-of-distributed from the training set, leading to failures on unseen attacks. In this paper, we tackle the failure on unseen jailbreaks problem, base on unsupervised latent direction discovery. We propose a bi-level adversarial training framework for zero-shot jailbreak defense. In the inner step, we simulate diverse jail-broken activations by extrapolating from refusal-state harmful-request activations via unsupervised latent direction discovery, which expands the coverage of real jailbreak activation subspaces. In the outer step, we train a potential-induced steering field to push these adversarial jailbroken states into refusal regions while keeping benign unchanged. Across three LLMs and six classical jailbreak families, our method achieves strong defense with attack success rates mostly below 5%, and rising subspace coverage throughout training helps explain the improved generalization.

Large language models (LLMs) are increasingly used as analyst assistants in security operations centers (SOCs), where they ingest log and alert data to produce triage labels, incident summaries, or remediation advice. We study a structural failure mode of this design: many log fields are attacker controlled. User agents, URLs, payloads, DNS queries, and attempted usernames can therefore carry instructions to the model alongside evidence of the intrusion. We call this setting log-substrate prompt injection. We introduce a four-class taxonomy of log-substrate attacks: direct override (S1), persona hijack (S2), context manipulation (S3), and obfuscated payloads (S4). We evaluate 48 strategy-defense-task combinations using gpt-4o-mini as the analyst. Three findings stand out. First, direct overrides are ineffective in our setting: all S1 classification attacks achieve 0% suppression. In contrast, persona hijacks suppress 68% of malicious logs under a naive classifier and remain effective under stronger defenses. Second, summarization is the highest-risk task: context manipulation reaches 96% injection success without defenses and 38% even with constrained output. Third, defenses reduce but do not eliminate the attack surface: average injection success falls from 26.6% under naive prompting to 11.8% under our strongest defense. We also compare empirical results to a deterministic mock analyst and find that simulation substantially mispredicts current model behavior, especially for direct overrides. These results suggest that SOC copilots should treat raw log content as adversarial input rather than ordinary analyst context.

Retrieval-augmented generation (RAG) has become central to large language model (LLM) deployments, grounding responses in enterprise or proprietary data to reduce hallucinations. However, this design introduces a new privacy risk: model outputs may signal the presence of specific documents in the retrieval corpus, enabling membership inference attacks (MIAs) that leak sensitive information. Existing MIAs are feasible, but they often rely on easily detected templated queries or require many non-templated yet costly and repetitive queries, limiting practicality. We ask: Can an adversary launch a limited-budget, surrogate-free, stealthy, and defense-agnostic membership inference attack using non-templated queries? We present MEntA (Membership Entailment Attack), a query-efficient MIA that leverages natural-language entailment to maximize information gained per query. By asking low-cost, broad, information-seeking questions and measuring entailment between model responses and candidate documents, MEntA eliminates the need for costly shadow models and large query budgets. Across NFCorpus, SCIDOCS, and TREC-COVID, MEntA achieves up to 0.991 AUC with only 5 queries, outperforming prior methods by up to 0.42 AUC under equivalent conditions. It remains effective under state-of-the-art (SOTA) RAG defenses, while current detectors either miss MEntA or flag benign queries at high rates. Regarding cost, MEntA reduces total attack cost by up to 65$\times$ lower compared to SOTA attacks under the same attack setting. Our findings expose the feasibility of realistic, low-cost privacy leakage in RAG systems and highlight the urgent need for privacy-aware retrieval and defense mechanisms.

We argue that LLM agent security is fundamentally an agent-human interaction (AHI) problem, not a purely algorithmic one. To substantiate this position, we conduct a systematic analysis of 59 academic papers, 21 production agent systems, and 26 security plugins as of April 2026. Our analysis reveals a striking pattern: the three widely deployed human-centric security mechanisms (policy specification, runtime approval, and scope configuration) dominate industry practice, each adopted by at least 14 of 21 systems (14, 15, and 16, respectively), while the categories most heavily studied in academia (intent anchoring and trust labeling) see zero production deployment. Yet current human participation mechanisms are far from satisfactory: they suffer from a fundamental trade-off between cognitive burden and security guarantees, leaving users caught between approval fatigue and uncontrolled agent autonomy. We make three contributions. First, through a systematic comparison of LLM-based and human-based intent alignment, we argue that human participation in agent security decisions is indispensable given current capabilities. Second, we quantify a pronounced industry-academia mismatch: the security mechanisms that practitioners actually deploy receive scant research attention, while the approaches that researchers favor remain undeployed. Third, we propose a three-direction research agenda and call for AHI security to be recognized as a first-class research citizen, one that demands its own design principles, evaluation methods, and theoretical foundations.

Privacy measurement instruments (e.g., CFIP, IUIPC, PAQ) predate GDPR by over a decade and measure privacy concerns, distinct from preferences for regulatory protections (e.g., data portability, erasure, automated decision-making rights). This leaves practitioners without tools to assess whether users value the GDPR mechanisms implemented in compliant policies. We developed a GDPR-grounded privacy preference measurement item bank by extracting 669 statements from all 99 GDPR articles, validated by: (1) two-round expert review achieving full consensus on accuracy, (2) semantic clustering into 10 parent themes and 87 subthemes, and (3) consensus review with 50 privacy experts (5 per theme) using a larger or equal than 4/5 vote retention threshold. The final 527-item bank comprises 9 parent themes and 73 subthemes (18 to 112 items per parent theme, 1 to 29 per subtheme), enabling targeted measurement across granularities while covering GDPR at mean pairwise expert agreement of approx. 85%. This work introduces a complementary measurement dimension aligning user preferences with regulatory mechanisms.

The development of brain-computer interfaces (BCIs) based on electroencephalograms (EEGs) has advanced significantly mainly to machine learning. Although the majority of earlier research has been on increasing classification accuracy, relatively little focus has been placed on security and robustness. According to recent research, EEG-based BCIs are susceptible to adversarial attacks, which can cause misdiagnosis due to minute, well-crafted disturbances. Evaluating model robustness against such perturbations is therefore critical for ensuring reliable deployment. In this study, we propose a lightweight custom Convolutional Neural Network (CNN) architecture to investigate adversarial robustness in EEG-based BCIs. The suggested method is assessed using two EEG datasets and contrasted with three novel CNN models tailored to EEG, namely EEGNet, DeepConvNet, and SleepEEGNet, under gradient-based adversarial attack scenarios. According to experimental findings, the suggested model continuously performs better in classification under adversarial perturbations compared to baseline models, indicating improved robustness. These findings highlight the potential of lightweight architectures for enhancing the reliability of EEG-based BCI systems under adversarial conditions.

Large language models (LLMs) are widely used for code generation, but their security reliability remains inconsistent across languages and prompting strategies. Existing prompt engineering improves functional correctness but rarely ensures consistent security outcomes. We introduce the Mitigation-Aware Chain-of-Thought (MA-CoT) framework, which embeds task-specific CWE mitigation guidance and language-aware safeguards to reduce recurring vulnerabilities in generated code. We evaluate MA-CoT across three LLMs (gpt-5, claude-4.5, gemini-2.5), three programming languages (C, Java, Python), and four prompting strategies (Vanilla, Zero-shot, CoT, MA-CoT) on a 200-task primary dataset, with external validation on LLMSecEval. Using static analysis with expert validation, MA-CoT reduces total security findings from 92 to 39 (57.6%) on the primary dataset and from 73 to 4 (94.5%) on LLMSecEval. High-severity findings (Blocker + Critical) drop from 90 to 39 (56.7%) and from 45 to 2 (95.6%), respectively. Across both datasets, MA-CoT is the only strategy that consistently improves security reliability; Zero-shot and CoT are less reliable and may increase vulnerability, especially in C. We further introduce a strict layered attribution of vulnerability drivers (language-core vs. stack layers) and show that residual risk concentrates in hardening-oriented patterns (e.g., OS- and toolchain-dependent), motivating secure-by-construction primitives alongside prompting.

The growing use of Large Language Models (LLMs) for automated code generation has enhanced software development efficiency, but often at the cost of security. Generated code frequently overlooks critical concerns, leaving it vulnerable to issues such as weak encryption and improper input validation. To investigate this problem, we present a comprehensive empirical evaluation of the security quality of LLM-generated code across five LLMs and four programming languages (Java, C++, C, and Python), examining the impact of multiple prompt engineering methods. We introduce a weaknesses-aware zero-shot chain-of-thought (WA-0CoT) prompting strategy that enriches prompts with security context using CWE mappings to guide model reasoning. Our empirical analysis, supported by chi-square tests, finds no statistically significant reductions in vulnerability frequency or density across prompt methods. However, prompting strategies, including WA-0CoT, systematically influence the compositional distribution of CWE categories, with effects varying by programming language. These findings suggest that while security-aware prompting alters the structure of generated weaknesses, prompt engineering alone is insufficient to reliably reduce overall vulnerability levels. The results highlight the importance of language-aware and model-aware prompt design when evaluating the security properties of LLM-generated code.

Android malware detectors often degrade after deployment because of concept drift, while full retraining at each maintenance step is costly. We propose a chronological adaptive maintenance framework that models deployment-time maintenance as a sequential decision problem. The framework learns a stable latent representation through self-supervised learning during initialization, freezes the encoder, measures latent drift in the fixed representation space, and performs lightweight downstream adaptation using a trainable adapter and classification head. A proximal policy optimization controller selects low-cost maintenance actions based on the detector state, including current utility, retention on a fixed memory set, latent drift indicators, and update cost. We evaluate the framework under a causal deployment-style protocol on emulator and real Android malware datasets with static and dynamic features. Results show that the RL controller provides a strong cost-aware adaptation strategy, consistently remaining among the top-performing policies while achieving a favorable balance between temporal performance, memory retention, and maintenance cost under non-stationary deployment conditions.

Sparse mixture-of-experts (MoE) language models activate only a small subset of parameters for each token, making router behavior a central part of model computation. This paper studies routing behavior of Mixtral 8x7B-Instruct under benign and harmful prompts using two complementary signals: activation-based routing scores derived from expert selection frequencies and gradient-based scores derived from router-gate sensitivities. We analyze expert- and layer-level routing behavior and conduct expert-suppression interventions. The results show that activation-based expert usage is broad and long-tailed, whereas gradient-based importance is concentrated. At expert level, benign and harmful prompt groups remain close under both signals with modest separation. At layer level, activation-based routing is most selective around layers 8-15, while gradient-based importance is concentrated in final layers. Expert classification shows most experts are shared across benign and harmful prompts, though a limited subset shows clear group preference. Top-ranked expert sets show stronger benign-malicious overlap under gradient scores than activation scores, suggesting concentration on a common late-layer expert set. In intervention experiments, suppressing top five benign-dominant experts from activation scores reduces restricted responses from 24 to 14 over 100 prompts, while suppressing gradient-derived experts reduces them from 34 to 22 with fewer unintended reversals. Overall, safety-relevant routing in Mixtral is subtle, depth-dependent, and distributed rather than dominated by a fixed set of experts.

The Model Context Protocol (MCP) standardizes how a large-language-model (LLM) agent and an external tool server exchange messages, but not trust: a host reads a server's self-declared tool list and dispatches calls, with no notion of which servers it may use, at what sensitivity, or which of a server's tools are in bounds. This work grew out of a concrete need – letting the Enclawed agent use Google's externally-operated MCP servers (Gmail, Calendar, Drive) safely, admitting the server and bounding the tools it may drive, without changing MCP or Enclawed's own tool application-programming interface (API). The mechanism we built, mcp-attested (shipped in both the open enclawed-oss distribution and the enclaved flavor), generalizes: the gap that makes an unmediated third-party connection unsafe for one user makes a regulated deployment impossible to accredit. We close it with three additive mechanisms: (1) a small, offline-signed clearance assertion a server publishes at a well-known Uniform Resource Identifier (URI) and a host verifies against a pinned trust root before any tool dispatch; (2) a deny-by-default per-server tool allowlist, so admitting a server is not trusting its every tool; and (3) a flavor-gated enforcement mode that turns the checks from warnings into hard denials, with every decision written to a tamper-evident audit log. We give the wire format, the verification algorithm, a security analysis, and an LLM-driven adversarial evaluation; we then state the design in normative Request-for-Comments (RFC 2119) form – schema, verification rules, error registry, well-known registration, and machine-checkable conformance vectors – so it can be adopted as an MCP addendum rather than reinvented. An unextended host ignores the well-known document and behaves exactly as today.

Deep-research agents are an alternative to conventional Web search. They use multi-agent pipelines to issue multiple Web searches related to user queries, retrieve relevant content from the answers, and generate detailed, evidence-based reports. We show that for many common search topics (including financial, medical, and product recommendations), agent-generated reports are consistently based on the same user-generated content (UGC) pages from platforms such as Reddit and Wikipedia. This retrieval overlap, combined with lax moderation and access controls for UGC, is an opportunity for subversion: an attacker who appends a short text to a single, frequently-retrieved page can cause agents to cite this text and promote attacker-chosen entities across many user questions. We evaluate this attack on representative deep-research systems across multiple topics, such as investment advice, antivirus software, restaurant recommendations, etc. Poisoning a single URL with as few as 13 words can be sufficient for the attacker's content to be retrieved in 57-76% of the agent executions for a given topic and cited in 38-51% of the generated reports. A more aggressive attack (poisoning an entire subreddit) achieves 30-53% citation rates even when the poison is only 0.5-4% of the retrieved content. The attacker does not need to know the phrasing of the user's question, nor the specific Web queries generated by the agent, nor the agent's internal retrieval and generation mechanisms. We then study defenses at different stages of the pipeline, including source-level filtering and output-based detection. Dropping UGC from retrieved content blocks the attack but degrades the quality of generated reports. We show that lightweight anomaly detection on inputs and outputs does not reliably identify poisoned content, nor the results of poisoning.

Many existing Artificial Intelligence (AI) solutions on mobile devices rely on an extensive collection of sensitive data, raising privacy concerns and often requiring storage for both context and model improvement. Apple's Private Cloud Compute (PCC) aims to address this by emphasizing mobile device integration and a privacy-first design. The central claim of PCC is that it does not store any user data and that user input and user accounts are unlinkable. While most of the PCC system specifications are public, compiled binaries add a layer of opaqueness. There are no reproducible builds, and there are no symbols within those binaries, creating potential discrepancies between the specification and what is shipped to the user. Additionally, the underlying models and interfaces for querying PCC are not openly accessible, limiting academic evaluation of model properties, such as accuracy. This poses a challenge in assessing whether a privacy-preserving approach like PCC is actually trustworthy while also providing high-quality answers. We are the first to reverse-engineer the PCC implementation on mobile devices to evaluate privacy aspects and to open its non-public interfaces on local devices to support custom PCC queries. We demonstrate this level of access beyond Apple's intended use cases by independently benchmarking the PCC model. We enable future research by making our PCC benchmarking framework publicly available.

Monitoring autonomous large language model (LLM) agents for covert malicious behavior is challenging due to delayed, context-dependent, and long-horizon attack patterns. Agents may pursue hidden objectives while maintaining superficially benign behavior, making detection difficult even with full trajectory access. Prior monitoring approaches improve scaffolding or ensemble aggregation, but treat each trajectory independently and do not learn from prior monitoring experience. Moreover, standard reasoning methods explain observed behavior without explicitly reasoning about agent beliefs, intentions, and goal alignment required to distinguish benign task execution from covert deviation. We propose Agent-ToM, a learning-to-monitor framework grounded in Theory-of-Mind (ToM) reasoning for security analysis of autonomous agents. Agent-ToM performs structured full-trajectory analysis by inferring beliefs, intent hypotheses with calibrated confidence, expected actions, and deviations from task-consistent behavioral baselines. At inference time, it employs a Reason-Verify-Refine pipeline to construct and validate monitoring decisions. At training time, Agent-ToM distills critique signals into a persistent semantic guardrail memory, enabling reusable belief- and intent-conditioned constraints across episodes. We evaluate Agent-ToM on adversarial agent monitoring benchmarks (SHADE-Arena and CUA-SHADE-Arena). Agent-ToM achieves strong precision-recall balance and outperforms state-of-the-art monitoring baselines, including ensemble methods, while using a coherent two-call reasoning pipeline. These results demonstrate that learning at the monitoring layer, combined with structured ToM reasoning and verification, provides an effective and deployable foundation for securing autonomous LLM agents.

Along with the recent rise in popularity of Electric Vehicles (EVs), Electric Vehicle Supply Equipment (EVSE) has emerged as a new target for cyber attacks. Therefore, ensuring the security and integrity of network communication between EVSE components and vehicular clients is a significant challenge that must be addressed. To this end, this paper proposes a Flow-based Analysis and Labeling for COnnected vehicular Network Cybersecurity (FALCON-C) framework. The FALCON-C framework leverages an autoencoder for anomaly detection and is trained on a small number of benign flows from the CICEVSE2024 dataset. The model's objective is to model benign flow behavior and identify malicious flows by detecting statistically different reconstruction error profiles. The results demonstrate that the model can successfully identify malicious flows, achieving 100% accuracy. Initially, some benign flows were misclassified as malicious, resulting in a suboptimal false positive rate. A thorough analysis of the autoencoder's performance and the nature of misclassified flows led to the development of a refined decision boundary, improving the framework's performance by 8.6%. FALCON-C is intended to support Security Operations Center activities by automating flow labeling, leading to the enhanced curation of reliable datasets that can be used for various activities, including threat modeling and hunting, decision auditing, and intrusion detection system refinement.

Electric Vehicles (EVs) have emerged as significant disruptors in the transportation sector over the past decade. Their growing popularity and adoption are accompanied by capital expenditures to deploy charging infrastructure. EV charging infrastructure sits at the intersection of the power grid, the network, and the vehicular client, creating an attractive surface for cyberattacks. Many machine learning-based cybersecurity countermeasures have been developed using various public and private datasets. These countermeasures, often intrusion detection systems, are limited in performance by the quality and expressivity of the training data. This work explores the most common datasets and modeling methods, identifies key limitations and open challenges, and proposes future directions to continue catalyzing innovation in the field. By addressing these data limitations, intrusion detection systems are better positioned to address the constantly evolving cyberthreat landscape of EV charging infrastructure.

Memorization in large language models has been studied almost exclusively through prefix-conditioned extraction, a natural choice for autoregressive models. However, diffusion language models (DLMs) can denoise masked tokens at arbitrary positions. Thus, prefix-only probing reveals only one facet of memorization in DLMs and significantly underestimates the risk of training-data extraction. In order to realistically model extractability of training data in DLMs, we introduce infilling extraction, a data-extraction protocol parameterized by an arbitrary binary mask that subsumes prefix-only probing and accounts for the bidirectional inductive bias of DLMs. Instantiating it on LLaDA-8B and Dream-7B across five extraction modes, three training pipelines, and three corpora covering verbatim and partial leakage, we find that mask geometry governs extractability: edge-conditioned masks extract up to three times more verbatim sequences than prefix-conditioned ones, and bidirectional access opens channels inaccessible in autoregressive models. In particular, we show that a realistic adversary with access to training data where personally identifiable information has been redacted, can even achieve higher recall on extracting redacted email addresses from DLMs than from scale-matched autoregressive models. Tunable parameters for decoding measurably affect extraction performance, while a follow-up supervised finetuning stage does not eliminate the prior memorization.