Journal of Hacking Techniques, Digital Crime Prevention and Computer Virology

Recent Advances in JPEG File Carving for Digital Forensics: A Systematic Review and Entropy-based Validation Framework

Mruganshi Patel, Vishvendu Bhatt — Mon, 30 Mar 2026 00:00:00 +0000

Recovering digital evidence is not as simple as scanning disk sectors anymore. The problem has become significantly more challenging over the last decade due to the evolution of storage architectures and the development of new compression methods. This paper reviews 30 peer- reviewed publications from 2009 to 2025, tracing the evolution of file carving through four stages—Syntax-driven (Generation I), Structure-aware (Generation II), Entropy-based (Generation III), and Hybrid AI-driven (Generation IV). While traditional carving tools only achieve a recovery rate of about 25.2% with fragmented data, which drops to almost zero with severe fragmentation, hybrid approaches using Extreme Learning Machines (ELM) and Generative Algorithms (GA) can achieve up to 97% recovery, although these results have limitations. To test these claims, the study developed the Entropy-Pillow Validation Framework and tested it on 143 forensic samples from NIST and Digital Corpora repositories, achieving an aggregate accuracy of 93.01% with perfect recall. It also proposes a dataset surrogacy metric, scoring 87.5%, to support the use of a public corpus instead of proprietary benchmarks. Finally, it examines two major threats to the field, the silent data erasure by SSD TRIM commands and the forensic challenges posed by the new JPEG AI standard.

DocSort AI: Document Governance System for Secure Business Records

Dipti Patil, Aditya Prabhash Lal, Harsh Santosh Patil, Karan Vijay Pendhari — Mon, 06 Apr 2026 00:00:00 +0000

In today’s fast-paced digital landscape, both individuals and enterprises struggle to manage an ever-growing volume of identity records, tax forms, and insurance policies across multiple uncoordinated platforms. This system was created to solve this fragmentation by providing an autonomous, highly secure document management ecosystem. Upon upload, the system utilises a local Vision Language Model (VLM) for zero-click categorisation, accurately identifying document types and extracting essential metadata (e.g., names, dates of birth, and unique identifiers) to route files properly into client-specific folders. To prevent cloud storage bloat without compromising text legibility, the platform utilises an adaptive, multi-tiered WebP compression algorithm, dynamically shrinking high-resolution PDFs and images to targeted kilobytes based on their initial size. Furthermore, it prioritises absolute data privacy through Envelope Encryption architecture backed by Google Cloud KMS, ensuring that every user's vault is secured by unique, KMS-wrapped Data Encryption Keys (DEKs). Together, these integrated systems deliver a highly scalable, duplicate-resistant, and confidential archiving solution designed to handle thousands of records natively.

Filename Spoofing Detection and Prevention: A Study

Sribhuvani J., Anand T. R. — Mon, 06 Apr 2026 00:00:00 +0000

Filename spoofing is an evasive technique used by attackers during file-based malware attacks to trick or deceive users by modifying file names, extensions, and encoding schemes. This technique takes advantage of operating systems and interface limitations to evade detection by pretending that a malicious executable file is a legitimate file type, such as a PDF file, image file, or text file. Techniques such as double extensions, hidden file extensions, Unicode homograph attacks, BIDI control character exploitation, and MIME type mismatches have greatly enhanced filename spoofing attacks. Microsoft Windows and Apple Mac operating systems are highly vulnerable to filename spoofing because they utilize filename extensions to identify file types. In the last ten years, filename spoofing has witnessed significant growth and development, including rule-based detection systems and sophisticated machine learning and behavioral analysis models. Security analysis services such as VirusTotal have proven that conventional signature-based antivirus systems are ineffective against filename spoofing because filename manipulation does not change the binary signature of a file. This has led to the development of MIME header verification, entropy-based anomaly detection, Unicode normalizations, and machine learning classifiers to detect and identify filename spoofing attacks effectively. This study provides a comprehensive overview of filename spoofing.

Ethical Implications of Artificial Intelligence in Social Media Platforms

Suraj R. Nalawade, Tapase H. O., Shravani Nalawade — Tue, 07 Apr 2026 00:00:00 +0000

The rapid integration of artificial intelligence (AI) in social media platforms has transformed the way individuals connect, share, and consume information. However, this advancement raises significant ethical questions that warrant critical analysis. This seminar explores the ethical implications of AI on social media, focusing on its potential benefits and the dangers it poses when improperly managed or utilized. Key concerns include data privacy breaches, algorithmic biases, and the amplification of misinformation, all of which can erode public trust and harm social well-being. Additionally, the seminar examines the psychological effects of AI-driven content personalization, which may reinforce echo chambers and impact mental health. Through a balanced discussion, this project underscores the necessity of developing robust ethical frameworks and regulatory measures to mitigate these risks. By fostering responsible AI practices, society can harness its potential for positive change while safeguarding against its adverse effects. This presentation aims to provide an in-depth understanding of the ethical challenges associated with AI on social media and offers actionable insights to promote accountability and transparency in its use.

Detectcy: ML-based Advanced Intrusion Detection System (IDS)

Nayan S. Bahirame, Aryan A. Bhosale, Prathamesh M. Mule, Prem P. Pardeshi — Tue, 07 Apr 2026 00:00:00 +0000

Cybersecurity risks have gotten more sophisticated and frequent, with the rise of digital communication and interconnected systems. Conventional intrusion detection systems (IDS) rely heavily on signature-based methodologies, limiting their ability to detect new or previously unknown threats. To overcome this drawback, this study presents Detectcy, an advanced IDS (software) that incorporates machine learning techniques to enable intelligent, adaptive, and more effective threat detection. Detectcy uses supervised learning algorithms like random forest, support vector machine (SVM), and decision tree to analyze network data and classify it as normal or malicious. To achieve reliable detection, the system comprises critical operations such as data gathering, preprocessing, feature extraction, and model training. Detectcy enhances efficiency and performance by utilizing feature selection and dimensionality reduction approaches. The proposed system is capable of detecting multiple types of cyber-attacks, including denial of service (DoS), Probe, Remote-to-Local (R2L), and User-to-Root (U2R) attacks. It also provides real-time monitoring and automated alert generation, enabling quick response to potential threats. Overall, Detectcy enhances detection accuracy, minimizes incorrect alerts while providing a flexible and scalable solution that can be effectively deployed in modern environments, including cloud-based platforms and IoT networks.