Sandeep Kumar Shukla

Rohit Kumar Sachan, Rachit Agarwal, Sandeep Kumar Shukla

The metadata aspect of Domain Names (DNs) enables us to perform a behavioral study of DNs and detect if a DN is involved in in-browser cryptojacking. Thus, we are motivated to study different temporal and behavioral aspects of DNs involved in cryptojacking. We use temporal features such as query frequency and query burst along with graph-based features such as degree and diameter, and non-temporal features such as the string-based to detect if a DNs is suspect to be involved in the in-browser cryptojacking. Then, we use them to train the Machine Learning (ML) algorithms over different temporal granularities such as 2 hours datasets and complete dataset. Our results show DecisionTrees classifier performs the best with 59.5% Recall on cryptojacked DN, while for unsupervised learning, K-Means with K=2 perform the best. Similarity analysis of the features reveals a minimal divergence between the cryptojacking DNs and other already known malicious DNs. It also reveals the need for improvements in the feature set of state-of-the-art methods to improve their accuracy in detecting in-browser cryptojacking. As added analysis, our signature-based analysis identifies that none-of-the Indian Government websites were involved in cryptojacking during October-December 2021. However, based on the resource utilization, we identify 10 DNs with different properties than others.

Vikas Maurya, Rachit Agarwal, Saurabh Kumar et al.

Due to the importance of Critical Infrastructure (CI) in a nation's economy, they have been lucrative targets for cyber attackers. These critical infrastructures are usually Cyber-Physical Systems (CPS) such as power grids, water, and sewage treatment facilities, oil and gas pipelines, etc. In recent times, these systems have suffered from cyber attacks numerous times. Researchers have been developing cyber security solutions for CIs to avoid lasting damages. According to standard frameworks, cyber security based on identification, protection, detection, response, and recovery are at the core of these research. Detection of an ongoing attack that escapes standard protection such as firewall, anti-virus, and host/network intrusion detection has gained importance as such attacks eventually affect the physical dynamics of the system. Therefore, anomaly detection in physical dynamics proves an effective means to implement defense-in-depth. PASAD is one example of anomaly detection in the sensor/actuator data, representing such systems' physical dynamics. We present EPASAD, which improves the detection technique used in PASAD to detect these micro-stealthy attacks, as our experiments show that PASAD's spherical boundary-based detection fails to detect. Our method EPASAD overcomes this by using Ellipsoid boundaries, thereby tightening the boundaries in various dimensions, whereas a spherical boundary treats all dimensions equally. We validate EPASAD using the dataset produced by the TE-process simulator and the C-town datasets. The results show that EPASAD improves PASAD's average recall by 5.8% and 9.5% for the two datasets, respectively.

Haoran Xi, Minghao Shao, Kimberly Milner et al.

Large language models are rapidly changing how learners acquire and demonstrate cybersecurity skills. However, when human--AI collaboration is allowed, educators still lack validated competition designs and evaluation practices that remain fair and evidence-based. This paper presents a cross-regional study of LLM-centered Capture-the-Flag competitions built on the Cyber Security Awareness Week competition system. To understand how autonomy levels and participants' knowledge backgrounds influence problem-solving performance and learning-related behaviors, we formalize three autonomy levels: human-in-the-loop, autonomous agent frameworks, and hybrid. To enable verification, we require traceable submissions including conversation logs, agent trajectories, and agent code. We analyze multi-region competition data covering an in-class track, a standard track, and a year-long expert track, each targeting participants with different knowledge backgrounds. Using data from the 2025 competition, we compare solve performance across autonomy levels and challenge categories, and observe that autonomous agent frameworks and hybrid achieve higher completion rates on challenges requiring iterative testing and tool interactions. In the in-class track, we classify participants' agent designs and find a preference for lightweight, tool-augmented prompting and reflection-based retries over complex multi-agent architectures. Our results offer actionable guidance for designing LLM-assisted cybersecurity competitions as learning technologies, including autonomy-specific scoring criteria, evidence requirements that support solution verification, and track structures that improve accessibility while preserving reliable evaluation and engagement.

Bikash Saha, Sandeep Kumar Shukla

Organizational cybersecurity policies are often examined to determine whether they adequately comply standard security controls. This task is difficult because control statements are abstract, whereas policy documents describe governance practices in varied natural language. As a result, policy-based control assessment is time-consuming, difficult to standardize, and often difficult to document in a traceable manner. To address this gap, we present PROPARAG, an audit support approach for evaluating organizational cybersecurity policies against security controls autonomously. For each control, the approach retrieves relevant policy evidence, assesses coverage, identifies missing elements, and generates supporting explanations and recommendations. We evaluate PROPARAG on two real-world organizational policy corpora using 1,007 NIST SP 800-53 controls across both closed-source and open-source large language models (LLMs). The framework achieves F1 scores of 88.54 on OrgA and 82.31 on OrgB. The evaluation also shows that PROPARAG identifies relevant gaps in documented organizational policies and generates grounded recommendations for each identified gap. This research provides foundation for LLM-powered autonomous control-level assessment of organizational cybersecurity policies.

Bikash Saha, Sandeep Kumar Shukla

Modern cybersecurity requires systematic ways to evaluate how detection systems respond to evolving and previously unseen attack behaviors. Existing malware repositories largely capture known patterns and provide limited support for stress-testing defenses against novel threats. To address this, we present MalGEN, a modular testbed that models adversarial workflows and generates executable artifacts in a controlled environment. The framework decomposes high-level attack objectives into structured stages, enabling the synthesis of diverse and multi-stage behaviors. We evaluate MalGEN across 1,920 benchmark settings covering multiple platforms and behavioral objectives, resulting in 977 executable samples. Analysis shows that the generated artifacts exhibit a wide range of malicious techniques and multi-stage attack patterns. However, 45.71% of these samples remain undetected by existing detection engines, which reveals notable gaps in current defenses. These findings provide practical insights into the limitations of widely used detection approaches and support the development of more robust security evaluation and testing practices.

Minghao Shao, Haoran Xi, Nanda Rani et al.

Large Language Model (LLM) agents can automate cybersecurity tasks and can adapt to the evolving cybersecurity landscape without re-engineering. While LLM agents have demonstrated cybersecurity capabilities on Capture-The-Flag (CTF) competitions, they have two key limitations: accessing latest cybersecurity expertise beyond training data, and integrating new knowledge into complex task planning. Knowledge-based approaches that incorporate technical understanding into the task-solving automation can tackle these limitations. We present CRAKEN, a knowledge-based LLM agent framework that improves cybersecurity capability through three core mechanisms: contextual decomposition of task-critical information, iterative self-reflected knowledge retrieval, and knowledge-hint injection that transforms insights into adaptive attack strategies. Comprehensive evaluations with different configurations show CRAKEN's effectiveness in multi-stage vulnerability detection and exploitation compared to previous approaches. Our extensible architecture establishes new methodologies for embedding new security knowledge into LLM-driven cybersecurity agentic systems. With a knowledge database of CTF writeups, CRAKEN obtained an accuracy of 22% on NYU CTF Bench, outperforming prior works by 3% and achieving state-of-the-art results. On evaluation of MITRE ATT&CK techniques, CRAKEN solves 25-30% more techniques than prior work, demonstrating improved cybersecurity capabilities via knowledge-based execution. We make our framework open source to public https://github.com/NYU-LLM-CTF/nyuctf_agents_craken.

Minghao Shao, Nanda Rani, Kimberly Milner et al.

Recent advances in LLM agentic systems have improved the automation of offensive security tasks, particularly for Capture the Flag (CTF) challenges. We systematically investigate the key factors that drive agent success and provide a detailed recipe for building effective LLM-based offensive security agents. First, we present CTFJudge, a framework leveraging LLM as a judge to analyze agent trajectories and provide granular evaluation across CTF solving steps. Second, we propose a novel metric, CTF Competency Index (CCI) for partial correctness, revealing how closely agent solutions align with human-crafted gold standards. Third, we examine how LLM hyperparameters, namely temperature, top-p, and maximum token length, influence agent performance and automated cybersecurity task planning. For rapid evaluation, we present CTFTiny, a curated benchmark of 50 representative CTF challenges across binary exploitation, web, reverse engineering, forensics, and cryptography. Our findings identify optimal multi-agent coordination settings and lay the groundwork for future LLM agent research in cybersecurity. We make CTFTiny open source to public https://github.com/NYU-LLM-CTF/CTFTiny along with CTFJudge on https://github.com/NYU-LLM-CTF/CTFJudge.

Nanda Rani, Divyanshu Singh, Bikash Saha et al.

The rise in cybercrime and the complexity of multilingual and code-mixed complaints present significant challenges for law enforcement and cybersecurity agencies. These organizations need automated, scalable methods to identify crime types, enabling efficient processing and prioritization of large complaint volumes. Manual triaging is inefficient, and traditional machine learning methods fail to capture the semantic and contextual nuances of textual cybercrime complaints. Moreover, the lack of publicly available datasets and privacy concerns hinder the research to present robust solutions. To address these challenges, we propose a framework for automated cybercrime complaint classification. The framework leverages Hinglish-adapted transformers, such as HingBERT and HingRoBERTa, to handle code-mixed inputs effectively. We employ the real-world dataset provided by Indian Cybercrime Coordination Centre (I4C) during CyberGuard AI Hackathon 2024. We employ GenAI open source model-based data augmentation method to address class imbalance. We also employ privacy-aware preprocessing to ensure compliance with ethical standards while maintaining data integrity. Our solution achieves significant performance improvements, with HingRoBERTa attaining an accuracy of 74.41% and an F1-score of 71.49%. We also develop ready-to-use tool by integrating Django REST backend with a modern frontend. The developed tool is scalable and ready for real-world deployment in platforms like the National Cyber Crime Reporting Portal. This work bridges critical gaps in cybercrime complaint management, offering a scalable, privacy-conscious, and adaptable solution for modern cybersecurity challenges.

Meet Udeshi, Minghao Shao, Haoran Xi et al.

Large Language Models (LLMs) have been used in cybersecurity such as autonomous security analysis or penetration testing. Capture the Flag (CTF) challenges serve as benchmarks to assess automated task-planning abilities of LLM agents for cybersecurity. Early attempts to apply LLMs for solving CTF challenges used single-agent systems, where feedback was restricted to a single reasoning-action loop. This approach was inadequate for complex CTF tasks. Inspired by real-world CTF competitions, where teams of experts collaborate, we introduce the D-CIPHER LLM multi-agent framework for collaborative CTF solving. D-CIPHER integrates agents with distinct roles with dynamic feedback loops to enhance reasoning on complex tasks. It introduces the Planner-Executor agent system, consisting of a Planner agent for overall problem-solving along with multiple heterogeneous Executor agents for individual tasks, facilitating efficient allocation of responsibilities among the agents. Additionally, D-CIPHER incorporates an Auto-prompter agent to improve problem-solving by auto-generating a highly relevant initial prompt. We evaluate D-CIPHER on multiple CTF benchmarks and LLM models via comprehensive studies to highlight the impact of our enhancements. Additionally, we manually map the CTFs in NYU CTF Bench to MITRE ATT&CK techniques that apply for a comprehensive evaluation of D-CIPHER's offensive security capability. D-CIPHER achieves state-of-the-art performance on three benchmarks: 22.0% on NYU CTF Bench, 22.5% on Cybench, and 44.0% on HackTheBox, which is 2.5% to 8.5% better than previous work. D-CIPHER solves 65% more ATT&CK techniques compared to previous work, demonstrating stronger offensive capability.

Swaroop Dora, Deven Lunkad, Naziya Aslam et al.

The rapid advancement of Large Language Models (LLMs) has enhanced software development processes, minimizing the time and effort required for coding and enhancing developer productivity. However, despite their potential benefits, code generated by LLMs has been shown to generate insecure code in controlled environments, raising critical concerns about their reliability and security in real-world applications. This paper uses predefined security parameters to evaluate the security compliance of LLM-generated code across multiple models, such as ChatGPT, DeepSeek, Claude, Gemini and Grok. The analysis reveals critical vulnerabilities in authentication mechanisms, session management, input validation and HTTP security headers. Although some models implement security measures to a limited extent, none fully align with industry best practices, highlighting the associated risks in automated software development. Our findings underscore that human expertise is crucial to ensure secure software deployment or review of LLM-generated code. Also, there is a need for robust security assessment frameworks to enhance the reliability of LLM-generated code in real-world applications.

Deepesh Chaudhari, Rachit Agarwal, Sandeep Kumar Shukla

The temporal aspect of blockchain transactions enables us to study the address's behavior and detect if it is involved in any illicit activity. However, due to the concept of change addresses (used to thwart replay attacks), temporal aspects are not directly applicable in the Bitcoin blockchain. Several pre-processing steps should be performed before such temporal aspects are utilized. We are motivated to study the Bitcoin transaction network and use the temporal features such as burst, attractiveness, and inter-event time along with several graph-based properties such as the degree of node and clustering coefficient to validate the applicability of already existing approaches known for other cryptocurrency blockchains on the Bitcoin blockchain. We generate the temporal and non-temporal feature set and train the Machine Learning (ML) algorithm over different temporal granularities to validate the state-of-the-art methods. We study the behavior of the addresses over different time granularities of the dataset. We identify that after applying change-address clustering, in Bitcoin, existing temporal features can be extracted and ML approaches can be applied. A comparative analysis of results show that the behavior of addresses in Ethereum and Bitcoin is similar with respect to in-degree, out-degree and inter-event time. Further, we identify 3 suspects that showed malicious behavior across different temporal granularities. These suspects are not marked as malicious in Bitcoin.

Mayank Pandey, Rachit Agarwal, Sandeep Kumar Shukla et al.

In cryptocurrency-based permissionless blockchain networks, the decentralized structure enables any user to join and operate across different regions. The criminal entities exploit it by using cryptocurrency transactions on the blockchain to facilitate activities such as money laundering, gambling, and ransomware attacks. In recent times, different machine learning-based techniques can detect such criminal elements based on blockchain transaction data. However, there is no provision within the blockchain to deal with such elements. We propose a reputation-based methodology for response to the users detected carrying out the aforementioned illicit activities. We select Algorand blockchain to implement our methodology by incorporating it within the consensus protocol. The theoretical results obtained prove the restriction and exclusion of criminal elements through block proposal rejection and attenuation of the voting power as a validator for such entities. Further, we analyze the efficacy of our method and show that it puts no additional strain on the communication resources.

Banwari Lal, Rachit Agarwal, Sandeep Kumar Shukla

The decentralization, redundancy, and pseudo-anonymity features have made permission-less public blockchain platforms attractive for adoption as technology platforms for cryptocurrencies. However, such adoption has enabled cybercriminals to exploit vulnerabilities in blockchain platforms and target the users through social engineering to carry out malicious activities. Most of the state-of-the-art techniques for detecting malicious actors depend on the transactional behavior of individual wallet addresses but do not analyze the money trails. We propose a heuristics-based approach that adds new features associated with money trails to analyze and find suspicious activities in cryptocurrency blockchains. Here, we focus only on the cyclic behavior and identify hidden patterns present in the temporal transactions graphs in a blockchain. We demonstrate our methods on the transaction data of the Ethereum blockchain. We find that malicious activities (such as Gambling, Phishing, and Money Laundering) have different cyclic patterns in Ethereum. We also identify two suspicious temporal cyclic path-based transfers in Ethereum. Our techniques may apply to other cryptocurrency blockchains with appropriate modifications adapted to the nature of the crypto-currency under investigation.

Rachit Agarwal, Tanmay Thapliyal, Sandeep Kumar Shukla

Smart Contracts (SCs) in Ethereum can automate tasks and provide different functionalities to a user. Such automation is enabled by the `Turing-complete' nature of the programming language (Solidity) in which SCs are written. This also opens up different vulnerabilities and bugs in SCs that malicious actors exploit to carry out malicious or illegal activities on the cryptocurrency platform. In this work, we study the correlation between malicious activities and the vulnerabilities present in SCs and find that some malicious activities are correlated with certain types of vulnerabilities. We then develop and study the feasibility of a scoring mechanism that corresponds to the severity of the vulnerabilities present in SCs to determine if it is a relevant feature to identify suspicious SCs. We analyze the utility of severity score towards detection of suspicious SCs using unsupervised machine learning (ML) algorithms across different temporal granularities and identify behavioral changes. In our experiments with on-chain SCs, we were able to find a total of 1094 benign SCs across different granularities which behave similar to malicious SCs, with the inclusion of the smart contract vulnerability scores in the feature set.

Rohit Kumar Sachan, Rachit Agarwal, Sandeep Kumar Shukla

The rise in the adoption of blockchain technology has led to increased illegal activities by cyber-criminals costing billions of dollars. Many machine learning algorithms are applied to detect such illegal behavior. These algorithms are often trained on the transaction behavior and, in some cases, trained on the vulnerabilities that exist in the system. In our approach, we study the feasibility of using metadata such as Domain Name (DN) associated with the account in the blockchain and identify whether an account should be tagged malicious or not. Here, we leverage the temporal aspects attached to the DNs. Our results identify 144930 DNs that show malicious behavior, and out of these, 54114 DNs show persistent malicious behavior over time. Nonetheless, none of these identified malicious DNs were reported in new officially tagged malicious blockchain DNs.

Bhaskar Pratim Mukhoty, Vikas Maurya, Sandeep Kumar Shukla

The energy output a photo voltaic(PV) panel is a function of solar irradiation and weather parameters like temperature and wind speed etc. A general measure for solar irradiation called Global Horizontal Irradiance (GHI), customarily reported in Watt/meter$^2$, is a generic indicator for this intermittent energy resource. An accurate prediction of GHI is necessary for reliable grid integration of the renewable as well as for power market trading. While some machine learning techniques are well introduced along with the traditional time-series forecasting techniques, deep-learning techniques remains less explored for the task at hand. In this paper we give deep learning models suitable for sequence to sequence prediction of GHI. The deep learning models are reported for short-term forecasting $\{1-24\}$ hour along with the state-of-the art techniques like Gradient Boosted Regression Trees(GBRT) and Feed Forward Neural Networks(FFNN). We have checked that spatio-temporal features like wind direction, wind speed and GHI of neighboring location improves the prediction accuracy of the deep learning models significantly. Among the various sequence-to-sequence encoder-decoder models LSTM performed superior, handling short-comings of the state-of-the-art techniques.