Papers

Today's young adults are most immersed in technology, leading in feelings of powerlessness in managing online privacy across many platforms, and particularly susceptible to phishing attacks. This raises questions about their general, wide-ranging attitudes towards and management of cybersecurity. How do young, tech-savvy adults approach cybersecurity? We seek a better understanding of their cybersecurity knowledge, attitudes and experiences, in particular in addressing deceptive online communications. We surveyed a group of `lead users': computing university students (n = 236). By combining thematic analysis of open-ended responses with quantitative data, we provide insights into their experiences and perceptions. While students demonstrate reasonable cybersecurity awareness, their cybersecurity experiences vary, and inconsistencies exist around their practices, perceptions of responsibility, and support structures. Findings also reveal four key thematic tensions: 1) Computing students are knowledgeable yet have persistent incorrect beliefs, 2) They learn more about keeping safe from sources outside the classroom, 3) They have limited assistance and have fallen victim to cybercrime, and 4) Many are confident, yet others are concerned about their own safety and responsibility. Through cluster analysis of attitudes, we identify two groups, with one feeling less prepared, less confident, yet expressing a desire to learn more. Established measures of intentions and objective knowledge were correlated to preparedness. Self-efficacy correlated to confidence and predicted cluster membership.

AI systems are increasingly evaluated in bounded environments that combine isolation, simulation, instrumentation, supervision, and evidence capture. For physical AI, AIoT, and cyber-physical systems, this shift is not a matter of terminology: the system under test may sense, decide, actuate, communicate, and fail through physical processes, networked devices, and human operators. This article develops an assurance-oriented account of AI sandboxes as controlled environments for testing, evaluation, verification, and validation across digital AI, embodied autonomy, and cyber-physical deployments. We formalize the sandbox boundary and a weakest-link rule for composing per-dimension evidence into a bounded deployment claim; separate major sandbox archetypes; define a cyber-physical threat model that includes attacks on the assurance apparatus itself; and introduce a measurement framework spanning fidelity, controllability, observability, containment, reproducibility, and governance artifacts, instantiated on three worked case studies of real sandboxes. The resulting threat model, taxonomy, and measurement framework clarify what a sandbox can validly test, which risks it can contain, and what forms of evidence it can support for safety, security, and regulatory assurance.

Domain-camouflaged injection attacks embed malicious instructions in retrieved content using domain-appropriate vocabulary, evading standard detectors that rely on syntactic injection markers. When detection fails, practitioners need to know which defense architectures reduce attack success. We evaluate five prompting-based defenses (spotlighting, paraphrasing, prompt sandwiching, and two combinations) against domain-camouflaged injection across three model families (Claude Haiku, Llama 3.1 8B, Gemini 2.0 Flash) and three deployment domains (financial, legal, general) using 3,510 trials. Paraphrasing retrieved content before agent processing is the most consistently effective defense in this benchmark, reducing camouflage attack success rate by 55-84% depending on model, and achieves lower attack success rates than our Llama Guard 4 configuration on every model tested. Defense effectiveness is strongly model-dependent: spotlighting halves attack success on Claude Haiku but provides no benefit on Llama 3.1 8B. Financial domain deployments face the highest residual risk at 26-33% baseline attack success rate, with no prompting-based defense fully eliminating the threat on weaker models. These results provide the first systematic evaluation of prompting-based defenses specifically against camouflage-class injection attacks and establish benchmark-based recommendations for practitioners. All tasks use synthetically constructed professional documents; whether these benchmark rankings generalize to real enterprise documents remains an open question.

The Mixed-Radix One-Time Pad (MR-OTP) extends the classical OTP to heterogeneous alphabets while preserving perfect secrecy. We provide a practical, bias-free method to convert raw binary entropy from a QKD source into uniform mixed-radix keys by identifying Horner's method and its inverse as the natural mapping between binary integers and mixed-radix tuples. We show that naive modular reduction induces bias and prove that rejection sampling restores uniformity with optimal expected cost. We establish end-to-end information-theoretic security for single and multi-session pipelines, quantify efficiency gains, present a batched extractor, and give unconditional and conditional results on the Base Recovery Problem.

Face Presentation Attack Detection (PAD) systems constitute a critical security layer in biometric authentication; however, existing approaches exhibit systematic performance disparities across demographic groups, disproportionately affecting individuals with darker skin tones. This paper presents a comparative empirical investigation of whether Vision Transformer architectures reduce demographic bias in face PAD systems relative to convolutional baselines. Experiments are conducted on the CASIA-SURF Cross-Ethnicity Face Anti-Spoofing (CeFA) dataset. Three architectures are evaluated: a Multimodal ViT-Tiny trained from scratch, a ResNet18 CNN baseline, and a pretrained DeiT-S fine-tuned on CeFA across African, East Asian, and zero-shot Central Asian demographic groups. DeiT-S achieves the highest overall accuracy of 97.27% and the lowest EER of 0.86%, outperforming ResNet18 at 90.15% accuracy. In terms of fairness, DeiT-S reduces the inter-ethnic ACER gap between African and East Asian subjects to 0.13%, compared to 0.75% reported in an LBP-based work [6], representing an 83% reduction. Most notably, while ResNet18 records a BPCER of 10.44% on zero-shot Central Asian subjects, DeiT-S maintains 2.89% on the same unseen group, demonstrating a 3.6x generalization advantage. These results suggest that pretrained Vision Transformers achieve superior PAD accuracy, produce smaller demographic performance gaps, and generalize more equitably across unseen demographic groups, indicating that cross-demographic fairness in PAD may partly be influenced by architectural design.

Retrieval-augmented generation (RAG) allows large language models to access external and private corpora for factual, domain-specific responses. Modern RAG pipelines use hierarchical navigable small world (HNSW) vector databases for efficient similarity search. When a user requests data deletion, the systems typically only mark the record as deleted, leaving the embedding on disk physically unchanged. This soft-delete operation raises compliance concerns under data-erasure and retention requirements such as GDPR Article 17 and HIPAA. Analysis on three HNSW implementations confirms that deleted vectors remain physically recoverable by accessing the raw index files at the storage layer, bypassing API access. Using the Vec2Text inversion model without domain-specific fine-tuning, we show this vulnerability on multiple real-world datasets and data modalities. On Wikipedia biographical living persons dataset (BLP), we successfully recover 25.5% of exact person names and 46.4% of geographic locations (ROUGE-L 0.185). Recovery reaches 100% for both patient age and gender markers (ROUGE-L 0.290) on highly structured, sensitive data (NIH Synthea dataset). On soft-deleted image embeddings, we show 100% tissue classification on histopathology patches (p=1.02e-07) and top-1 identity recovery reaches 99% on facial embeddings (p<0.01). This work introduces Epoch Key Rotation, which encrypts vectors and discards the key upon deletion. Epoch key rotation reduces observed PII recovery to 0% and completes in 2.5 ms for 500 deleted vectors (approximately 0.005 ms/record). Additionally, it generates an ECDSA-signed cryptographic proof as an auditable record of the deletion event.

Statistical watermarks help organizations attribute large language model (LLM) outputs, yet existing detectors often struggle when watermark signals are weak, texts are repetitive, or watermarks are edited. We propose signature filtering, a detection-time module that enhances watermark detection without modifying watermark embedding and text generation. It learns a small set of “signature” tokens whose presence makes watermark tests unreliable, and removes these tokens before detection. The signatures are obtained by solving a mixed-integer linear program on a small training set, with constraints that maximize the true positive rate. We additionally derive finite-sample and asymptotic bounds under several attacker models (color-blind, color-adaptive, and distributionally correlated). On four well-known watermark families (Kgw, Sweet, Unigram, Exp), four benchmark corpora (C4, MBPP, HumanEval, Code-Search-Net), and six LLMs (Opt-1.3b, Opt-6.7b, Llama2-13b, Llama3.1-8b, Qwen2.5-14b, Phi-3-medium-14b), 2- and 3-gram signatures raise detection rates in weak-signal and low-entropy settings from 8~31% without filtering to 78~99% with filtering, while keeping false positives controllable and often negligible. In stress tests where we scramble sentences and perturb 25~50% of tokens by dilution, deletions, and substitutions, 2-gram filters for Kgw-style watermarks preserve most of the clean-text detection gains, often matching or outperforming the advanced WinMax watermark detector. Signature filtering thus provides a simple, scalable, and model-agnostic add-on to strengthen watermark-based provenance checks for LLM text in information processing workflows.

In China, a burgeoning underground market sells citizens subscription-based censorship circumvention proxies known as ”airports”. We present the first systematic study of this ecosystem, combining user surveys, social media analysis, and active network measurements. We find that airports are by far the most popular off-the-shelf censorship circumvention tool in China, used by over half of our 1,667~survey respondents, who cite their ease of use, performance, and access to geo-restricted services like ChatGPT and Netflix. By scanning the Internet and scraping Telegram announcement channels, we identify 3,431 active airports built on a handful of open-source toolkits. We subscribe to 35 airports and characterize their performance, which often surpasses direct connections through the Great Firewall due to a distinctive multi-hop architecture. However, airports also pose new challenges and security risks: they accept payment through commercial services like Alipay, suffer frequent government takedowns, and are difficult for clients to configure optimally. Many airports also deploy their own distinct censorship policies. Airports are far more widely used than other circumvention tools from the academic literature, but introduce new forms of fragility and control, offering both lessons and opportunities for future circumvention research.

Testing implementations of binary protocols for specification compliance requires inputs that satisfy both structural and semantic constraints. Purely random generation and primitive mutations are often insufficient for exploring semantically meaningful behaviors in protocols that rely on Type-Length-Value (TLV) encoding, yet domain-specific compliance testing tools require deep protocol expertise and significant manual effort to construct. This work investigates whether grammar-level mutation combined with LLM-based code synthesis can serve as a viable, more generalizable approach to specification compliance testing. We evaluate the approach on PKCS#1 v1.5 signature verification – a widely deployed TLV-encoded standard with a formally verified testing oracle (Morpheus) – across 48 cryptographic library implementations. We reproduced 10 of 13 non-trivial specification violation categories previously identified by Morpheus, including all 5 signature forgery categories, and discovered 1 previously unreported discrepancy. We found that LLM hallucination – occurring in 82.5% of generated scripts – is the primary factor limiting effectiveness, not the mutation strategies. We identify five distinct hallucination types and show that their distribution varies systematically across mutation categories: structural mutations are implemented with 13.3% fidelity while constraint mutations achieve 30.3% correctness but suffer the highest rate of mutations being fully ignored (8.1%). These findings reveal a striking gap between operational reliability (99.8%) and semantic fidelity (17.5%), providing actionable guidance on when LLM-based code synthesis can be trusted in specification-driven testing pipelines.

Large foundation models deployed on third-party and shared accelerator infrastructure face a practical risk of model exfiltration that existing defenses do not fully address. In common serving deployments, model providers control the VM or bare-metal serving stack but not the surrounding hardware substrate. The host to GPU interconnect, accelerator fabric, and neighboring infrastructure components remain outside the tenant's trust boundary and have been shown to be exploitable. Hermes demonstrates lossless DNN reconstruction from passive PCIe observation, while TunnelS exfiltrates HBM contents at high throughput via driver-level access without disrupting inference. Co-tenant VMs can further access memory-mapped interfaces or misconfigured RDMA regions without physical co-location. These attacks exploit a common property of ML systems: model weights are stored in large, contiguous, and repeatedly accessed memory regions, making intercepted PCIe transfers and HBM dumps rich enough to reveal model structure and parameters. We present CloakLM, a software-only memory-obfuscation framework that removes this structural regularity without changing the inference stack's logical view of memory. CloakLM combines three mechanisms: PCIe traffic shaping, inter- and intra-layer weight shuffling, and physical HBM page remapping. Authorized execution retains a valid virtual memory layout with negligible overhead, while unauthorized observers see fragmented and semantically incoherent state. CloakLM integrates with vLLM and PyTorch, requires no hardware changes, and complements confidential computing. Evaluation on distributed inference workloads using LLaMA and Qwen models shows near-native performance while significantly increasing resistance to PCIe snooping and HBM dump attacks, making inference-time model exfiltration substantially less practical.

Tool-using language-model agents introduce security failures that go beyond unsafe text: they can disclose protected objects, write persistent memory, send messages, modify databases, or trigger harmful code and tool effects. Existing evaluations often collapse these stages into a single attack success rate, making it difficult to tell whether a model merely agreed with an attacker or actually produced observable harm. We introduce SafeClawBench, a staged benchmark for tool-using agent security with 600 controlled adversarial tasks across six attack families: direct and indirect prompt injection, tool-return injection, memory poisoning, memory extraction, and ambiguity-driven unsafe inference. SafeClawBench reports three separate endpoints: semantic attack acceptance, audit-visible harm evidence, and sandbox-observed tool/state harm. Evaluating five agent endpoints under four prompt-level policies, we find that these endpoints capture different failure modes. Without additional prompt protection, semantic failure rates vary widely across models, from 9.0% to 44.2%. Audited harm evidence is narrower than semantic failure, and under a separate executable protocol some matched task identities produce sandbox harm despite passing the Semantic Core call: in a 12,000-row matched analysis, 291 of 347 observed sandbox harms occur in rows that pass the semantic check. Prompt policies change endpoint outcomes, but their effects depend on both model and protocol. SafeClawBench provides a reproducible framework for comparing agent models and prompt-policy conditions without conflating textual compliance, evidence-supported harm, and executable state changes. The open-source dataset is available at https://huggingface.co/datasets/sairights/safeclawbench.

Enterprise intrusion response still depends on static playbooks and analyst-driven triage, creating delay between alert generation and containment. We present Agentra, a supervisable multi-agent Intrusion Response System (IRS) framework that converts alerts from IDS, EDR, and XDR platforms into structured incident response plans grounded in MITRE ATT&CK, MITRE D3FEND, and NIST CSF 2.0. Agentra decomposes response reasoning across role-scoped agents, validates proposed plans through a bounded Planner–Validator review loop, screens retrieved threat intelligence through a Moderator security gateway, gates actions through an Action Catalog and risk score, and records decisions in an append-only audit log. We evaluate Agentra against a static OASIS CACAO v2.0 cyber-playbook baseline on a 120-event corpus drawn from ThreatHunter-Playbook, Splunk BOTSv3, and DARPA OpTC. The strongest configuration improves FP-aware IRS F1 from 0.61 to 0.84 and restores the projected harmful-action rate to the static baseline level of 0.0% after Planner-only configurations introduce unsafe overreaction. These results indicate that multi-agent response planning can improve ontology-grounded IRS coverage while preserving analyst approval and auditability.

Cybersecurity literature reviews require a reproducible denominator: the set of papers that a protocol includes before screening and synthesis begin. Today, that denominator is often reconstructed from publisher portals, bibliographic indices, and scholarly application programming interfaces (APIs) whose coverage, formats, and query semantics change over time. This paper presents TopVenues, an open-source system that materializes corpus construction as a versioned research artifact. TopVenues declares a venue and year scope, uses the DBLP Computer Science Bibliography (DBLP) as its primary metadata source, enriches records with abstracts and BibTeX entries via open scholarly APIs and publisher-specific extractors, and stores the results in a monotonic SQLite snapshot, accessible via a command-line interface (CLI), a web interface, and export paths for review workflows. The May 2026 snapshot contains 9,925 bibliographic records from 11 configured security-focused, security-relevant adjacent, and survey venues, with 99.86% abstract coverage and 99.99% BibTeX coverage; keyword search over the full corpus completes in under 31 ms, and a 252-test suite validates the data-integrity invariants. The fixed denominator also enables repeatable measurement: 29.2% of 2024 to 2025 papers from the four top-ranked security conferences in our scope appear as arXiv preprints, with a median of five months before publication, and a prior-author-track-record filter yields a 16.5x relative risk (2.5x conventional lift) at 90% recall for triaging preprints that later appear in the same venue set. TopVenues links corpus construction to auditable cybersecurity measurement by making the corpus itself executable, inspectable, and citable. The artifact is available at https://github.com/sidneibarbieri/topVenues.

Adversarial patches pose a practical threat to modern object detectors. Prior work shows vulnerability, but three gaps limit actionable insight: (i) few score-based black-box attacks jointly optimize patch location, texture, and size under tight query budgets; (ii) success is rarely tied to the patch's visual footprint; and (iii) evaluations often conflate EOT robustness with plain-view suppression. We present \method{}, a query-efficient, budget-adaptive black-box attack that couples a lightweight Contextual Thompson-Sampling placer with NES-style pixel updates, growing the patch only when progress stalls. Reporting is anchored by a strict plain-image suppression test; EOT is audited but never used as a substitute for success, and optional appearance/printability weights expose strength–visibility trade-offs. Across YOLOv5, Faster R-CNN, and YOLOS, \method{} achieves strong suppression on CNN-based detectors and substantial suppression on the transformer-based detector, using compact patches and exposing clear query–footprint trade-offs relative to fixed-size and heuristic baselines. A print–capture pilot further shows transfer across unseen physical objects and viewpoints.

Federated learning allows multiple clients to jointly train a shared model by sending gradient updates to a central server while keeping raw inputs local. However, prior gradient inversion attacks show that these updates can reveal enough information to reconstruct client inputs. Existing attacks on transformers either optimize dummy inputs to match the true client updates, which is costly and unstable for modern models, or exploit the low rank of attention gradients to identify a subspace containing the true layer embeddings, followed by a discrete membership test for candidate tokens. However, this token test is brittle under numerical noise, i.e., from quantization or Differential Privacy (DP), and scales poorly for encoder models with non-causal attention. We introduce TIGER, a continuous gradient inversion attack that turns this subspace signal into a differentiable objective. Instead of searching over tokens or matching full gradients, TIGER directly optimizes token embeddings to minimize their distance to the subspace. Our experiments demonstrate that on encoder-only models, TIGER substantially improves both reconstruction quality and runtime over existing attacks, while on decoder models, TIGER is more robust than prior subspace-based attacks, enabling the first successful reconstructions in DP-defended federated learning settings.

Injecting malicious knowledge into retrieval-augmented generation (RAG) systems can manipulate retrieved evidence and mislead downstream generation, posing a serious security threat for AI applications. Existing RAG injection attacks mainly rely on manipulating external knowledge bases, such as crafting malicious corpus. However, the synthetic text crafted by such data-centric methods could be detectable, leading to the failure of attacks. Beyond corpus manipulation, open-source retrievers are increasingly exposing RAG systems to model-centric attacks. In this paper, we propose conflict-aware retriever editing, i.e., CAREATTACK, a model-centric retriever attack framework for malicious knowledge injection in RAG. Specifically, CAREATTACK consists two stages of conflict-aware retriever editing and attack-preserving anchor repair. Conflict-aware retriever editing adapts efficient closed-form parameter editing to the dense retrieval model, promoting malicious knowledge above benign competing passages and resolving potential parameter conflicts through graph-based conflict detection and parameter editing projection. Then, attack-preserving anchor repair performs lightweight calibration on the edited retriever to further eliminate the impact on non-target prompts while preserving the attack effectiveness for target prompts. We instantiate CAREATTACK on Qwen3-Embedding-0.6B and BGE-M3, and conduct evaluation on three benchmark datasets. Experimental results demonstrate our method substantially promote malicious passages into the retrieved knowledge of RAG systems and can perform attacks for batches of target prompts and passages, given the access of retrieval model parameters. Since most RAG systems are built upon open-source retrieval models, this work reveals a practical attack surface in RAG systems. Codes are public accessible at https://anonymous.4open.science/r/CareAttack-3F1C.

With sophisticated cyber-attacks becoming increasingly prevalent, modern networks require intelligent autonomous cyber-defense agents trained via Reinforcement Learning (RL). These agents employ neurosymbolic approaches such as behavior trees with learning-enabled components (LECs) to learn, reason, adapt, and implement security rules while maintaining critical operations. However, these autonomous networks are partially observable systems, i.e., the cyber-attacker's (red agent's) actions are not observable, making it difficult for the defender to predict red actions, learn red policies, or assess the attacker's intrusion levels. To address this, we propose a Policy Learning Technique using imitation learning to learn policies for partially observable RL agents with discrete states and discrete actions. We apply this technique in an autonomous cyber environment to predict red agent's actions from network observations and defender actions. Integrated with a neurosymbolic cyber-defense agent, our method effectively handles different red policies and achieves high prediction accuracy across diverse simulated scenarios.

Consensus protocols form the core of blockchains and other replicated state machines, ensuring that all correct nodes process the same totally ordered log of input transactions. In fault-free executions, performance is driven by the good-case transaction latency – the time between a transaction becoming known to all nodes and its confirmation by the consensus protocol – which depends on both how frequently proposals are made and, once made, how quickly they are confirmed. While prior work has established tight lower bounds on confirmation latency that modern protocols already achieve, it remains open whether the inter-proposal time can be further reduced below the state-of-the-art of one network delay. We introduce Gatling, an atomic broadcast protocol that achieves arbitrarily small inter-proposal times under rotating leader schedules; in particular, smaller than the network delay. Gatling runs multiple parallel instances of a black-box atomic broadcast protocol and staggers their proposal schedules to generate proposals in faster succession than state-of-the-art protocols. A deterministic interleaving rule merges the outputs of these instances into a single global log. We analyze the effects of head-of-line blocking caused by crashed leaders, and derive Gatling's optimal number of parallel instances. We further study the impact of Gatling on predictable validity and present two variants that retain this property. Finally, our experiments confirm that Gatling can be used with off-the-shelf component protocols to achieve low latency without fine-tuning the component protocol for minimum latency.

Agent skills are emerging as an important attack surface in LLM-based systems. Through an empirical study of existing skill scanners, we find that current defenses primarily rely on textual descriptions, manifests, and source code as the main signals for security analysis, which can leave visually conveyed malicious intent insufficiently examined. This creates a practical blind spot: harmful operational instructions hidden in images may bypass scanning while still being recoverable by multimodal agents during deployment. To systematically investigate this threat, we propose SkillCamo, a document-mediated multimodal instruction attack that conceals malicious instructions within images bundled with a skill while rewriting the surrounding documentation to naturally reference those images as part of the normal workflow. Thus, the attack does not rely on the image alone, but on the joint interpretation of textual guidance and visual payload at execution time. To defend against such attacks, we further propose ExecScan, an execution-grounded multimodal scanning module that performs intent extraction, behavior reconstruction, abuse assessment, and deliberative execution simulation over skill artifacts. ExecScan jointly analyzes documentation, code, referenced resources, and visual content to recover hidden instructions, reconstruct executable behavior chains, and identify downstream risks such as exfiltration, destruction, persistence, deception, and privilege escalation. Extensive experiments show that image-hidden malicious instructions challenge existing skill scanners, while ExecScan can improve the skill scanning performance.

We evaluate the adversarial robustness of three frontier large language models (LLMs) developed by Anthropic, Opus 4.8, Fable 5 and its successor Fable 5.1, against four families of automated jailbreak attack across 7,826 harmful intents spanning a ten-category harm taxonomy. Using the HackAgent red-teaming framework, hundreds of thousands of adversarial attempts were generated and every apparent success was independently re-adjudicated by the same panel of five frontier judge models ($\geq 4/5$ agreement) for all three targets. All models resist the majority of attacks, but the residual surface is larger than aggregate framing suggests: it is dominated by adaptive iterative attacks, while static obfuscation is near-fully neutralised. On the two attack families run against all three targets (TAP and PAP), with identical denominators, the ordering does not follow release date: Fable 5 is most robust ($2.72\%$), Opus 4.8 follows ($5.76\%$) and the newest model, Fable 5.1, is least robust ($8.19\%$). The three fail in largely disjoint places, and conditioning on prompts each model actually answered shows why: adaptive tree search breaks all three at broadly similar rates ($10.1\%$, $9.7\%$, $12.1\%$), whereas one-shot persuasion separates them by more than an order of magnitude ($3.1\%$, $0.7\%$, $9.5\%$). What distinguishes these models adversarially is their resistance to framing, not to search. Even so, they produced 1,315 (Opus 4.8), 620 (Fable 5) and 1,282 (Fable 5.1) panel-confirmed harmful completions spanning every harm category, found automatically and cheaply, usually within one or two refinement steps, by an attacker model with no human expert in the loop. Even the best, most-tested frontier models remain reliably breakable under sustained automated pressure, and robustness does not monotonically improve from one version to the next.

Multi-stage cyberattacks span system, network, and browser logs. Detecting them requires correlating events across all three sources. Machine learning methods can learn these cross-source patterns, but they need labeled multi-source data. Existing public datasets fall short. Network-only datasets such as CICIDS and UNSW-NB15 miss host and browser activity. Host-focused datasets such as LMDG and CICAPT-IIoT lack browser telemetry. ATLAS includes all three sources but labels events only as malicious or benign, without MITRE Adversarial Tactics, Techniques, and Common Knowledge (ATT&CK) technique granularity. No public dataset combines all three sources with per-entry ATT&CK technique labels. We close the gap by building a multi-source log dataset of 870 sessions (70 attack, 800 benign) and approximately 2.3 million events. We captured system, network, and browser activity simultaneously on Windows endpoints. We labeled malicious events with ATT&CK technique IDs, covering 12 tactics and 53 techniques. We generated all attack data using real tools, including Remote Access Trojan (RAT), Command and Control (C2) tunnels, and cloud exfiltration. To demonstrate learnability, we fine-tuned three Small Language Models (SLMs) (Qwen2.5-1.5B, Llama-3.2-3B, Phi-4-Mini) using Low-Rank Adaptation (LoRA). We compared each against its base variant across ten metrics on two tasks: chunk classification and ATT&CK technique identification. Fine-tuning improved every model on every metric. Chunk classification accuracy rose from approximately 8% in the base variants to between 90% and 97% after fine-tuning. Technique identification remained challenging, with the best exact-match accuracy at 42%, although high partial-match scores show the models captured most of the underlying reasoning.

Classifying Cyber Threat Intelligence (CTI) using MITRE Adversarial Tactics, Techniques, and Common Knowledge (ATT&CK) is essential for proactive defense, but historically required extensive human effort. Pre-Large Language Model (LLM) automation sped up this process, but could not resolve the complex language and multi-step attack patterns found in unstructured CTI reports. LLMs addressed previous limitations by using contextual reasoning to understand unstructured text. However, current evaluations rely on simplified, single-technique sentences that ignore the complexity of real-world CTI reports, which often leads to inflated performance results. Consequently, the baseline performance of open-source LLMs on complex unstructured CTI reports remains unevaluated. To address this gap, we constructed a ground-truth dataset of 2,076 human-annotated sentences (1,281 technique-positive, 795 negative) from 83 complex unstructured CTI reports. These sentences were mapped to 114 unique ATT&CK techniques using a six-phase annotation process, achieving ąppa} = 0.68 inter-annotator agreement. Using this dataset, we evaluated seven open-source LLMs ranging from 8B to 236B parameters across prompt strategy and temperature configurations. The highest-performing LLM achieved a micro-averaged F1 score of 0.22, establishing the empirical baseline for multi-label ATT&CK classification on complex unstructured CTI. Parameter size showed a statistically significant positive correlation with F1 score. Prompt strategy and temperature produced no statistically significant gains across model configurations. These results indicate that current open-source LLMs are insufficient for production-grade ATT&CK classification. The dataset, benchmark, and findings provide a reproducible foundation for future CTI research.

Large language model applications build prompts from templates, and Handlebars is a widely used templating engine and the default prompt-template format in Microsoft Semantic Kernel. Its double-brace {{x}} expression HTML-escapes the interpolated value and is documented as the safe default; its triple-brace {{{x}}} expression inserts the value raw. We show that this choice silently governs an application's exposure to structural role injection, where attacker-controlled data carries chat role delimiters that forge a higher-privilege turn. A model-free analysis establishes the mechanism: Handlebars escaping rewrites angle brackets but not square brackets, colons, or Markdown hashes, so it neutralises ChatML, Llama-3, and XML role delimiters (survival rate 0.00) while leaving Llama-2 [INST], legacy Human:/Assistant:, and Markdown ### delimiters intact (survival rate 1.00 for the last two). We then run 5760 trials across seven delimiter families, two attack objectives, and four models (GPT-3.5 Turbo, GPT-4o mini, GPT-4.1 mini, Claude Haiku 4.5) at a combined API cost of 1.63 USD. GPT-3.5 Turbo follows the task-hijack instruction in 97% of raw and 91% of escaped trials, with the escaping protection concentrated in the angle-bracket families and absent for the colon- and Markdown-based families; the harder secret-exfiltration objective, which does not saturate, exposes the same family interaction more cleanly. Claude Haiku 4.5 resists both objectives almost entirely. The escaped default protects only the delimiter schemes whose characters HTML escaping happens to cover, gives no protection for the rest, and cannot substitute for a structural separation of instruction and data.

Encrypted control can preserve the privacy of data and parameters while the necessary computations can be outsourced to a cloud server. To ensure the integrity of the received values from the cloud, i.e., that they have not been changed, however, strong assumptions or verification algorithms are needed. Previous methods require computationally expensive cryptographic protocols or are only applicable to static computations. In this paper, we present a novel type of verification algorithm for linear dynamic encrypted control. We utilize system-theoretic input-output properties of the controller for artificial challenge signals, which are processed in the cloud in parallel with the requested control input, to check the correctness of the results at the plant. This results in almost no additional computational load, wrong computations are revealed with high probability, and no replay attacks are possible.

Large language models (LLMs) are widely used to fulfill users' information needs; users ask LLMs about the weather, pose educational questions, and consult them for legal assistance. One particularly understudied area is digital security and privacy (S&P), where users may seek LLMs' help on how to secure their online accounts or protect their computers from cyber attacks. To the best of our knowledge, no prior study has collected or analyzed the S&P questions users ask LLMs; prior research on LLM response quality relied on expert-authored S&P misconceptions or FAQs rather than user queries. Drawing from WildChat, a dataset of 3.2M user-LLM conversations collected in the wild, our study identifies 14,727 S&P prompts and categorizes them into nine categories covering a wide range of S&P topics. From the S&P prompts, we sampled 450 and performed a thematic analysis to characterize the S&P questions users ask LLMs. Separate from the thematic analysis, we curated 270 advice-seeking S&P prompts, where users ask for recommendations, guidance, or specific S&P information. We measured LLM response quality and consistency when posing the prompt to LLMs 10 times. We found that commercial LLMs outperform open-weight models (GPT 5.5 provided "good enough" responses on 98% of prompts; Llama 4 on 47%). However, among prompts that received high-quality responses on average, commercial models sometimes produce contradictory responses across runs, risking confusing or misleading users.

Generative AI models lower the bar for content creation, making it easy for any user to create professional-looking images, text and music with minimal effort. This has enabled a new cottage industry around creation of "AI slop" mass quantities of mediocre content produced to generate revenue, often through misrepresentation as human-authored content, or scams involving automated scripts and fake consumption. While there are obvious parallels between the AI-slop industry and "traditional" email spam networks, it might be too early to determine if AI slop generation can grow into a similar self-sustaining industry. In this paper, we look specifically at the music industry, and explore the question: Can we prevent AI music slop from growing into a self-sustaining shadow industry? To answer this question, we characterize the current state of AI slop in music, and its pipeline from generation, distribution, and consumption by users on streaming platforms. By examining growth and engagement on Spotify, we confirm that AI music exhibits AI slop characteristics: the overwhelming majority (93%) of AI music receive few, if any listener plays, and are rarely recommended. AI musicians "spray and pray," releasing large volumes of music across multiple genres in hopes of generating a hit. We also explore the AI slop pipeline by generating and publishing our own AI tracks onto streaming through 11 indie music distributors. We find distributors have inconsistent and largely unenforced policies on AI music, making it surprisingly easy to publish mass produced AI songs. Finally, we consider AI music detection, and find that current methods lack accuracy or robustness. As generation costs decrease, we believe slop generation in music will become self-sustainable, unless concrete steps are taken by the music industry. We consider and discuss potential mitigation methods based on our findings.

We study the privacy of releasing functional posterior sample paths from a Gaussian process (GP) when the entire training set including covariates and responses is private. Unlike standard differential-privacy (DP) mechanisms that inject external noise, posterior sampling is intrinsically random and we show that this randomness provides useful privacy guarantees. We derive Rényi-DP guarantees separating privacy leakage through the posterior mean from a distinct channel induced by the data-dependent posterior covariance. The analysis identifies effective ridge regularisation and covariance scale as the principal privacy-controlling quantities and yields sharper guarantees in several regimes of practical interest as well as extensions to repeated and adaptive releases. Membership inference attacks confirm the predicted dependence on regularisation, covariance scale and the number of released paths. Utility experiments on downstream posterior sampling tasks identify noisy observation regimes where privacy-compatible regularisation preserves useful samples. Finally we identify large-data asymptotic regime in which the privacy parameter and posterior mean-square risk vanish simultaneously, yielding privacy for free. Together, these results provide a comprehensive characterisation of privacy and utility of GP posterior sampling.

NFV/SDN orchestration lets operators instantiate and steer traffic through virtual firewalls, IDS/IPS replicas, WAF clusters, zero-trust gateways, backup inspection paths, and migration targets on demand. Operators often treat these options as monotone improvements: more inspection capacity, lower nominal latency, or broader placement flexibility should not degrade the service. That intuition can fail even when the new option is locally attractive. We study a security-induced Braess paradox in service function chain (SFC) orchestration, where adding a defensive option worsens the post-adaptation equilibrium by concentrating traffic and adversarial value on shared security resources. We define Braessian security-management actions, derive a sufficient condition for paradox emergence under affine load-dependent VNF delay, and give a pre-deployment orchestration screen that rejects, caps, or reserves harmful options. A multi-tenant SFC experiment suite applies the model to four topology-derived settings: a fat-tree datacenter, NSFNET-style WAN, GEANT-style WAN, and edge/fog topology. Under default parameters in the Braessian regime identified by the theory, naive defensive expansion raises equilibrium service cost by 27.2-30.8% and increases risk concentration by factors of 6.1-9.7. Paradox-aware constrained use keeps the residual penalty below 1.9%, reduces service cost by 20.0-22.1% relative to naive expansion, and lowers a concentration-sensitive attack-loss proxy by 93.5% on average.

Cyber deception and Moving Target Defense are promising strategies that aim to disrupt adversaries by increasing uncertainty. However, sustaining long-lived, credible interactive sessions with adversaries remains an open challenge. Large Language Models (LLMs) offer a promising path toward more dynamic deception systems, but suffer from key limitations that fundamentally limit their applicability, including: lack of persistent state, output inconsistencies, hallucinations, latency, and susceptibility to behavioral subversion that may reveal the deception. We propose ShellGames, an SSH shell simulator based on LLM designed to address these limitations. ShellGames combines five complementary techniques: (i) Automatic Chain-of-Thought and few-shot learning to improve correctness; (ii) memory management to maintain system state coherency; (iii) speculative command execution to reduce response latency; (iv) smart routing of complex interactive commands to a sandboxed environment; and (v) subversion detection leveraging the constrained input-output domain of shell environments. To enable systematic evaluation, we introduce a standardized benchmarking protocol and dataset spanning correctness, consistency, state tracking, and robustness tasks. ShellGames achieves $0.898$ command accuracy on correctness ($+5.3pp$ over baselines), $0.918$ sequence-level accuracy on consistency ($+36pp$), $0.98$ state tracking accuracy ($+18.3pp$), and $0.95$ accuracy on robustness ($+37pp$). A user study with $n=20$ participants confirms that ShellGames achieves realism comparable to a real shell under free exploration and outperforms traditional honeypots on perceived command coverage.

Digital technologies are now central to children's learning, play, communication, identity formation, and social participation. Yet dominant approaches to children's online safety often rely on containment mechanisms, including bans, age gates, parental controls, monitoring, and screen-time restrictions. These approaches can be useful in specific contexts, but they often frame child protection primarily as a problem of restricting access to systems designed for adults. In this paper, we argue that this framing is inadequate for children's digital lives and insufficient as a security paradigm. We propose Child-fit security, a design paradigm in which technologies likely to be used by children treat a child as legitimate users, not attackers to be excluded, vulnerabilities to be patched, or risks to be managed. In this paradigm, children's wellbeing, development, privacy, safety, agency, and rights become core security requirements. This shifts the focus of protection from apps, accounts, and data to the child-system relationship, which means protecting both the child and their participation. We conceptualise child-fit security, contrast it with containment-oriented approaches, define its core principles, and discuss its implications for security design. We conclude by presenting a research agenda for making child-fit security operational.