Evolving attacker capabilities demand realistic and continuously updated cyberattack emulation for threat-informed defense and security benchmarking. Towards automated attack emulation, this paper defines modular attack actions and a linking model to organize and chain heterogeneous attack tools into causality-preserving cyberattacks. Building on this foundation, we introduce Aurora: an automated cyberattack emulation system powered by symbolic planning and large language models (LLMs). Aurora crafts actionable, causality-preserving attack chains tailored to Cyber Threat Intelligence (CTI) reports and target environments, and automatically executes these emulations. Using Aurora, we generated an extensive cyberattack emulation dataset from 250 attack reports, 15 times larger than the leading expert-crafted dataset. Our evaluation shows that Aurora significantly outperforms existing methods in creating actionable, diverse, and realistic attack chains. We release the dataset and use it to evaluate three state-of-the-art intrusion detection systems, whose performance differed notably from results on older datasets, highlighting the need for up-to-date, automated attack emulation.

Cyber threat intelligence (CTI) is increasingly consumed not by human analysts but by LLM agents that compose multi-step investigations at query time. The harness side of this shift has matured rapidly (planning loops, tool protocols, context management), but the corpus side has not: threat reports and vulnerability databases are still packaged for retrieval-augmented generation, as opaque chunks behind an embedding index. We argue that this substrate, not model capability, is the bottleneck on agentic CTI investigation, and present CTIFoundry, an agent-native corpus scaffold. At build time, CTIFoundry materializes the latent structure of a CTI corpus: a deterministic ontology graph over four authoritative knowledge bases (CVE, CWE, CAPEC, ATT CK) whose official cross-references become typed, traversable edges; a span-grounded report layer whose canonical, alias-resolved cross-vendor entities index provenance-carrying chunks; and hybrid dense+lexical retrieval surfaces. At query time this structure is exposed through seven typed tools and three procedural skills mounted on a stock open-source agent harness. On the public CTIConnect benchmark, swapping only the action surface lifts the identically-harnessed agent by +0.19 to +0.28 overall F1 across a four-model, two-provider panel: a small model on CTIFoundry surpasses a flagship on the flat substrate, and the gain is not bought with search effort, since on both Claude models the scaffolded agent is more accurate at roughly half the tool calls. An ablation attributes it: typed structure carries the larger share, procedural skills convert structure into discipline, and the two compose super-additively, because skills bind only to structure that exists.

Cloud-native applications have seen significant growth over the past decade due to their crucial role in bridging software development, deployment, and operations. The rapid pace of the container-based software supply chain, however, has introduced considerable security challenges. While Linux provides adequate privilege mechanisms that integrate with popular container runtimes like Docker, poor security practices such as granting excessive privileges to third-party containers remain prevalent. This issue persists largely because deriving appropriate privilege configurations for third-party container images is still an ad hoc process. To assign the correct privileges to a container, DevOps teams require precise knowledge of both the underlying system and the behavior of the containerized application. In this paper, we present ConfigWiz, a practical tool designed to automate the privilege configuration process for containerized applications, with a focus on the Docker platform. ConfigWiz establishes a detailed mapping between system calls (syscalls) and capabilities, incorporating accuracy enhancements through analysis of the Linux kernel source code. To address privilege requirements for auxiliary operations not required by the image itself, ConfigWiz leverages Large Language Models (LLMs) to analyze discussions from online forums, providing a flexible and secure privilege configuration process.

We propose CRAFT, a red-teaming alignment framework that leverages model reasoning capabilities and hidden representations to improve robustness against jailbreak attacks. Unlike prior defenses that operate primarily at the output level, CRAFT aligns large reasoning models to generate safety-aware reasoning traces by explicitly optimizing objectives defined over the hidden state space. Methodologically, CRAFT integrates contrastive representation learning with reinforcement learning to separate safe and unsafe reasoning trajectories, yielding a latent-space geometry that supports robust, reasoning-level safety alignment. Theoretically, we show that incorporating latent–textual consistency into GRPO eliminates superficially aligned policies by ruling them out as local optima. Empirically, we evaluate CRAFT on multiple safety benchmarks using two strong reasoning models, Qwen3-4B-Thinking and R1-Distill-Llama-8B, where it consistently outperforms state-of-the-art defenses such as IPO and SafeKey. Notably, CRAFT delivers an average improvement in reasoning safety and improvement in final-response safety over the base models, demonstrating the effectiveness of hidden-space reasoning alignment.

Kubernetes Operators, automated tools designed to manage application lifecycles within Kubernetes clusters, extend the functionalities of Kubernetes, and reduce the operational burden on human engineers. While Operators significantly simplify DevOps workflows, they introduce new security risks. In particular, Kubernetes enforces namespace isolation to separate workloads and limit user access, ensuring that users can only interact with resources within their authorized namespaces. However, Kubernetes Operators often demand elevated privileges and may interact with resources across multiple namespaces. This introduces a new class of vulnerabilities, the Cross-Namespace Reference Vulnerability. The root cause lies in the mismatch between the declared scope of resources and the implemented scope of the Operator logic, resulting in Kubernetes being unable to properly isolate the namespace. Leveraging such vulnerability, an adversary with limited access to a single authorized namespace may exploit the Operator to perform operations affecting other unauthorized namespaces, causing Privilege Escalation and further impacts. To the best of our knowledge, this paper is the first to systematically investigate Kubernetes Operator attacks. We present Cross-Namespace Reference Vulnerability with two strategies, demonstrating how an attacker can bypass namespace isolation. Through large-scale measurements, we found that over 14

As Advanced Persistent Threats (APTs) become more complex and destructive, attack investigation has gained importance. Analysts use provenance graphs for causality analysis on Point-Of-Interest (POI) events to capture critical events. However, existing methods suffer from problems of high false positives, high overhead, and high latency due to the vast size of the provenance graph and the rarity of critical events. We propose SParse for constructing critical component graphs (i.e., consisting of critical events) from streaming logs in real time. Our approach is based on two key observations: 1) Critical events exist in suspicious semantic graphs (SSGs) composed of interaction flows between suspicious entities, and 2) Information flows accomplishing the attacker's goal exist as paths. SParse uses a two-stage framework that first constructs the SSG using a state-based mode with semantic transfer rules and storage strategies. Then, it identifies suspicious flow paths (SFPs) related to the POI event and quantifies each path's influence to filter irrelevant events. Evaluation on a large-scale attack dataset shows our system generates a critical component graph (similar to 113 edges) in 1.6 seconds, which is 2014 & times; smaller than the backtracking graph (similar to 227,589 edges). It is also 25 & times; more effective in filtering irrelevant edges compared to other state-of-the-art techniques.

We propose **FROST**, an attention-aware method for efficient reasoning. Unlike traditional approaches, FROST leverages attention weights to prune uncritical reasoning paths, yielding shorter and more reliable reasoning trajectories. Methodologically, we introduce the concept of reasoning outliers and design an attention-based mechanism to remove them. Theoretically, FROST preserves and enhances the model’s reasoning capacity while eliminating outliers at the sentence level. Empirically, we validate FROST on four benchmarks using two strong reasoning models (Phi-4-Reasoning and GPT-oss-20B), outperforming state-of-the-art methods such as TALE and ThinkLess. Notably, FROST achieves an average **58.72%** reduction in token usage and a **10.64%** improvement in accuracy over the base model. Furthermore, in evaluations of attention outlier metrics, FROST reduces the maximum infinity norm $\lVert \mathbf{x} \rVert_{\infty}$ by **15.97%** and the average kurtosis by **91.09%** compared to the base model.

As cyber-attacks become increasingly sophisticated and stealthy, accurately distinguishing between benign behavior and malicious intrusions has become both more critical and more challenging. Provenance-based intrusion detection systems (PIDS) show strong potential for detecting malicious activities through fine-grained causality analysis, which has gained significant attention from both industry and academia. Among the various PIDS approaches, rule-based systems are particularly favored for their low overhead, real-time detection capability, and interpretability. However, these systems face challenges in reducing false positive rates, primarily due to the lack of fine-tuned rules and specific environments. In this paper, we introduce CAPTAIN+, a rule-based PIDS that autonomously adapts to diverse environments online. Specifically, we propose three adaptive parameters to adjust the detection configuration for nodes, edges, and alarm generation thresholds. Initially, we build a differentiable tag propagation framework and utilize the gradient descent algorithm to optimize these adaptive parameters based on the training data. In this extended version, we integrate an online learning module into the detection stage to dynamically optimize adaptive parameters based on real-time feedback from the detection process. We evaluate CAPTAIN+ based on data from DARPA TC, OpTC datasets, and PKU ASAL datasets. The results demonstrate that CAPTAIN+ offers superior detection accuracy, lower detection latency, reduced runtime overhead, long-term resilience against concept drift, and more interpretable detection results compared to state-of-the-art PIDS.

Microservices are popular and widely used in the cloud. However, realizing cost-effective and high-performance microservice orchestration is challenging for Cloud Service Providers (CSPs). Current orchestration mechanisms have limited flexibility and resource efficiency in scheduling and would cause sluggishness in scaling, which brings unnecessary costs to CSP. This paper presents PROPHET, a microservice orchestrator for optimizing service scheduling and scaling. To improve scheduling flexibility and resource utilization, we propose a ranking-based p-batch scheduling mechanism, which adopts a pairwise ranker to obtain resource-efficient scheduling plans for large-scale microservice applications rapidly. To advance the scaling agility, we design a proactive prediction-based scaling mechanism, which performs scaling in advance based on resource usage prediction. Our evaluations are conducted on a real-world cluster with the public Alibaba cluster dataset and datasets collected from the cluster. The results indicate that PROPHET can significantly reduce the number of nodes running in the cluster and improve scaling. This shows great potential in achieving cost-effective and high-performance microservice orchestration.

Detecting advanced persistent threats (APTs) at a host via data provenance has emerged as a valuable yet challenging task. Compared with attack rule matching, machine learning approaches offer new perspectives for efficiently detecting attacks by leveraging their inherent ability to autonomously learn from data and adapt to dynamic environments. However, the scarcity of APT samples poses a significant limitation, rendering supervised learning methods that have demonstrated remarkable capabilities in other domains (e.g., malware detection) impractical. Therefore, we propose a system called TAGAPT, which is able to automatically generate numerous APT samples with provenance-level granularity. First, we introduce a deep graph generation model to generalize various graph structures that represent new attack patterns. Second, we propose an attack stage division algorithm to divide each generated graph structure into stage subgraphs. Finally, we design a genetic algorithm to find the optimal attack technique explanation for each subgraph and obtain fully instantiated APT samples. Experimental results demonstrate that TAGAPT can learn from existing attack patterns and generalize to novel attack patterns. Furthermore, the generated APT samples 1) exhibit the ability to help with efficient threat hunting and 2) provide additional assistance to the state-of-the-art (SOTA) attack detection system (Kairos) by filtering out 73% of the observed false positives. We have open-sourced the code and the generated samples to support the development of the security community.

In this study, we applied random regression test-day model for genomic prediction in the Holstein population in Shandong Province of China with respect to different reference populations, using either 150 k chip genotypes or imputed sequence genotypes. Three different reference populations were considered, i.e., the Shandong (SD) reference population consisting of 1 688 Holstein cows from Shandong Province, the Non-SD reference population consisting of 5 299 Holstein cows from other parts of China, and the combined population of the two. The SD reference resulted in higher prediction accuracy than the Non-SD reference, although the former was much smaller than the latter. The combined reference further increased the accuracy. These results indicate that the accuracy of genomic prediction cross-population within breed is low, even though the reference population is large. Using imputed sequence data may not significantly improve the cross-population prediction ability. However, the inclusion of data from other populations into the reference population can improve the accuracy of genomic selection.

Advanced Persistent Threat (APT) attacks have caused significant damage worldwide. Various Endpoint Detection and Response (EDR) systems are deployed by enterprises to fight against potential threats. However, EDR suffers from high false positives. In order not to affect normal operations, analysts need to investigate and filter detection results before taking countermeasures, in which heavy manual labor and alarm fatigue cause analysts miss optimal response time, thereby leading to information leakage and destruction. Therefore, we propose Endpoint Forecasting and Interpreting (EFI), a real-time attack forecast and interpretation system, which can automatically predict next move during post-exploitation and explain it in technique-level, then dispatch strategies to EDR for advance reinforcement. First, we use Cyber Threat Intelligence (CTI) reports to extract the attack scene graph (ASG) that can be mapped to low-level system logs to strengthen attack samples. Second, we build a serialized graph forecast model, which is combined with the attack provenance graph (APG) provided by EDR to generate an attack forecast graph (AFG) to predict the next move. Finally, we utilize the attack template graph (ATG) and graph alignment plus algorithm for technique-level interpretation to automatically dispatch strategies for EDR to reinforce system in advance. EFI can avoid the impact of existing EDR false positives, and can reduce the attack surface of system without affecting the normal operations. We collect a total of 3,484 CTI reports, generate 1,429 ASGs, label 8,000 sentences, tag 10,451 entities, and construct 256 ATGs. Experimental results on both DARPA Engagement and large scale CTI dataset show that the alignment score between the AFG predicted by EFI and the real attack graph is able to exceed 0.8, the forecast and interpretation precision of EFI can reach 91.8

Security practitioners face growing challenges in exploit assessment, as public vulnerability repositories are increasingly populated with inconsistent and low-quality exploit artifacts. Existing scoring systems, such as CVSS and EPSS, offer limited support for this task. They either rely on theoretical metrics or produce opaque probability estimates without assessing whether usable exploit code exists. In practice, security teams often resort to manual triage of exploit repositories, which is time-consuming, error-prone, and difficult to scale. We present AEAS, an automated system designed to assess and prioritize actionable exploits through static analysis. AEAS analyzes both exploit code and associated documentation to extract a structured set of features reflecting exploit availability, functionality, and setup complexity. It then computes an actionability score for each exploit and produces ranked exploit recommendations. We evaluate AEAS on a dataset of over 5,000 vulnerabilities derived from 600+ real-world applications frequently encountered by red teams. Manual validation and expert review on representative subsets show that AEAS achieves a 100

Long-term geographic isolation and breeding programs both influence population characteristics. Leiqiong cattle, a native breed from the southernmost region of China, are renowned for disease and heat resistance, with two subgroups on Hainan Island and the Leizhou Peninsula. However, the genomic differences between them remain unexplored. In this study, we conducted genomic comparisons using whole-genome sequencing data from the two subgroups of Leiqiong cattle and three commercial breeds to assess their population structures. Leiqiong cattle in Hainan exhibited lower genetic diversity and a pure ancestral content due to their isolation from the mainland. In contrast, the subgroup in Guangdong displayed higher genetic diversity and mixed ancestry, influenced by the intrusion of commercial breeds. The genetic divergence between them was evaluated by estimating a genetic distance of 0.08 and a split time of 3400 to 4250 years ago, highlighting the role of geographical barriers in speciation. Notably, two candidate genes were identified through selection sweeps, including PIP4K2A, potentially related to immunity, and TNFSF4, possibly involved in hair follicle development. Our findings reveal the different genetic structures and genomic characteristics in the two subgroups of Leiqiong cattle, providing valuable insights into their evolutionary history and establishing a foundation for future breeding strategies.

Provenance-based intrusion detection is an increasingly popular application of graphical machine learning in cybersecurity, where system activities are modeled as provenance graphs to capture causality and correlations among potentially malicious actions. Graph Neural Networks (GNNs) have demonstrated strong performance in this setting. However, traditional statically-provisioned GNN inference architectures fall short in meeting two crucial demands of intrusion detection: (1) maintaining consistently low detection latency, and (2) handling highly irregular and bursty workloads. To holistically address these challenges, we present GraphFaaS, a serverless architecture tailored for GNN-based intrusion detection. GraphFaaS leverages the elasticity and agility of serverless computing to dynamically scale the GNN inference pipeline. We parallelize and adapt GNN workflows to a serverless environment, ensuring that the system can respond in real time to fluctuating workloads. By decoupling compute resources from static provisioning, GraphFaaS delivers stable inference latency, which is critical for dependable intrusion detection and timely incident response in cybersecurity operations. Preliminary evaluation shows GraphFaaS reduces average detection latency by 85

While penetration testing plays a vital role in cybersecurity, achieving fully automated, hands-off-the-keyboard execution remains a significant research challenge. In this paper, we introduce the "Planner-Executor-Perceptor (PEP)" design paradigm and use it to systematically review existing work and identify the key challenges in this area. We also evaluate existing penetration testing systems, with a particular focus on the use of Large Language Model (LLM) agents for this task. The results show that the out-of-the-box Claude Code and Sonnet 4.5 exhibit superior penetration capabilities observed to date, substantially outperforming all prior systems. However, a detailed analysis of their testing processes reveals specific strengths and limitations; notably, LLM agents struggle with maintaining coherent long-horizon plans, performing complex reasoning, and effectively utilizing specialized tools. These limitations significantly constrain its overall capability, efficiency, and stability. To address these limitations, we propose CHECKMATE, a framework that integrates enhanced classical planning with LLM agents, providing an external, structured "brain" that mitigates the inherent weaknesses of LLM agents. Our evaluation shows that CHECKMATE outperforms the state-of-the-art system (Claude Code) in penetration capability, improving benchmark success rates by over 20

Serverless computing, also known as Function-as-a-Service (FaaS), is widely used in modern applications. Function instances share the underlying physical infrastructure, which makes co-location attacks possible and leads to the leakage of sensitive information such as function activity states. Existing work has respective limitations in serverless scenarios because of incomplete detection coverage, long training time, and intrusion into the function's runtime environment. In this paper, we propose FAASGUARD, an obfuscation framework to protect function activity states in network side-channels and enhance privacy in serverless applications. To be specific, we design an adaptive obfuscation strategy selection mechanism to make FAASGUARD flexible. We design a traffic camouflage method to make obfuscated traffic indistinguishable from normal traffic, making FAASGUARD invisible. In order not to affect normal traffic, we propose a tag-based obfuscation mechanism to identify obfuscated packets. The preliminary evaluation results show that FAASGUARD can conceal function activity states with negligible resource overhead.

Recent advances in Large Language Models (LLMs) have led to impressive alignment where models learn to distinguish harmful from harmless queries through supervised finetuning (SFT) and reinforcement learning from human feedback (RLHF). In this paper, we reveal a subtle yet impactful weakness in these aligned models. We find that simply appending multiple end of sequence (eos) tokens can cause a phenomenon we call context segmentation, which effectively shifts both harmful and benign inputs closer to the refusal boundary in the hidden space. Building on this observation, we propose a straightforward method to BOOST jailbreak attacks by appending eos tokens. Our systematic evaluation shows that this strategy significantly increases the attack success rate across 8 representative jailbreak techniques and 16 open-source LLMs, ranging from 2B to 72B parameters. Moreover, we develop a novel probing mechanism for commercial APIs and discover that major providers such as OpenAI, Anthropic, and Qwen do not filter eos tokens, making them similarly vulnerable. These findings highlight a hidden yet critical blind spot in existing alignment and content filtering approaches. We call for heightened attention to eos tokens' unintended influence on model behaviors, particularly in production systems. Our work not only calls for an input-filtering based defense, but also points to new defenses that make refusal boundaries more robust and generalizable, as well as fundamental alignment techniques that can defend against context segmentation attacks.

Micro-segmentation isolates network segments within different parts of an application, reducing potential attack surfaces. This technique has become increasingly common for enhancing security in cloud application infrastructures. Despite its benefits, the complexity of managing numerous service interactions can make defining and maintaining micro-segmentation policies challenging and prone to errors. Previous solutions have attempted to simplify policy creation, but gaps remain in their applicability, auditability, and response times. In this paper, we proposed the first configuration-based approach, AugoSeg, which automates the generation of micro-segmentation policies for cloud-native applications. By analyzing network configurations in service containers, AugoSeg identifies service dependencies and automatically creates corresponding policies. This system specifically targets commonly used, behavior-focused configurations, addressing the shortcomings of earlier systems through its design. We systematically evaluated AugoSeg, using the 184 services from 61 popular projects, covering 14 programming languages. The results illustrated that AugoSeg can completely model service dependencies for over 96.7% of projects and formulate restrictive policies in an average time of 7.13 s. It effectively restricts attackers' lateral movements within networks. This evaluation not only underscores the efficiency of AugoSeg but also demonstrates its practical applicability in cloud environments, setting a new approach for micro-segmentation in cloud-native security.

Serverless computing, also known as Function-as-a-Service (FaaS), has gained popularity due to its flexibility, scalability, and transparent development. However, attacks against serverless are also increasing. Unfortunately, complex multi-layer FaaS architecture and frequently launched lightweight functions help attackers conceal their tracks. Specifically, 1) fully tracking the behavior of a function requires crossing multiple layers of FaaS. 2) Intrusive auditing components in functions affect function startup latency and performance. 3) Accurately provenance cross-layer function invocations require integrating data from multiple sources. In this paper, we propose FaaSTracker, a cross-layer, non-intrusive, efficient provenance framework for accurately tracking user function behaviors in FaaS. FaaSTracker tracks function behaviors across layers using a non-intrusive agent without any modifications to the function. In addition, it correlates data from multiple sources to construct a provenance graph of function workflows to locate attackers. We implement FaaSTracker on the OpenFaaS platform and evaluate its performance using real-world serverless applications. Compared with state-of-the-art serverless provenance systems, FaaSTracker provides a more accurate and complete view of provenance graphs and reduces 54.0% CPU and 48.9% memory resources.