Journal of Information Security System and Cyber Criminology Research

AI-Based Multimedia Authenticity Detection System (MMADS) for Combating Digital Misinformation

Pabbireddy Bhavya Sri Jyothi — 2026-10-05

The high rate of spreading deepfakes, AI-created images, synthetically cloned audio, phishing emails, and malicious URLs constitutes an increasing threat to the integrity of digital information, trust, and institutional security. Current detection systems are mostly unimodal, focusing on each type of content separately, and, therefore, do not offer full protection against the current multi-vector misinformation attacks. This study introduces the Multimedia Authenticity Detection System (MMADS), which is an integrated artificial intelligence system designed to identify counterfeited online content in five media types, which include pictures, videos, audio files, web addresses, and email messages. The MMADS architecture is divided into five functional layers, namely user interface, backend API, AI processing, decision output, and persistent storage, built upon a production-grade technology stack, which consists of React.js with TypeScript on the frontend, FastAPI with Uvicorn on the backend, a core detection engine based on TensorFlow and Keras CNN models, and Supabase with PostgreSQL to store results. Other libraries, such as OpenCV and Pillow to support visual preprocessing, Librosa and PyDub to extract audio features, and NumPy to perform numerical computations, are also used. Each detection model is trained on massive-scale data and, to guarantee data confidentiality and performance autonomy, the model is run in-house. In each of the media items under analysis, MMADS produces a binary decision, Authentic or Fake, and a confidence score, allowing checking the material with transparency and responsibility. Experimental analysis has shown that the system has an overall detection accuracy of 96.8 and an AUC-ROC of 0.972, and an end-to-end response latency of 142 milliseconds per media item. The main advance of this study is a platform for scalable, explainable, and computationally efficient multi-modal authenticity detection, which is compatible with the United Nations Sustainable Development Goal 16, which is digital trust, peace, and institutional integrity.

A Comprehensive Survey of Watermarking, Fingerprinting, and Blockchain-Based Authentication for Provenance Verification of AI-Generated Images

Sriharsha S. H. — 2026-09-21

The rapid growth of diffusion models and generative adversarial networks has increased the availability of AI-generated images and intensified the need to verify their provenance, including origin, authenticity, and modification history. This survey reviews 20 works published from 2023 to 2026 and organizes them into four primary provenance approaches: digital watermarking, perceptual hashing/fingerprinting, blockchain-anchored registries, and synthetic-media detection, while also examining recent watermark-removal and forgery attacks. The review covers diffusion-native methods such as Tree-Ring Watermarks and Stable Signature, post-hoc methods such as InvisMark and Watermark Anything, perceptual-fingerprinting approaches such as DinoHash, and blockchain-based registry mechanisms. It further analyzes attacks including MarkSweep, Warfare, the Next-Frame Prediction Attack, and boundary-leakage attacks. A recurring limitation is that provenance verification is often reduced to an authenticity or match decision, even when the underlying watermark can carry multi-bit information; this limits direct assessment of the type and severity of image modifications. Other challenges include cross-architecture generalization, spatial redundancy, inconsistent threat models, and the absence of a universally adopted evaluation benchmark. The survey synthesizes these findings and identifies research directions for building more robust, interpretable, and practically deployable AI image provenance systems.

Hybrid CNN–Transformer Model for Real-Time Cybercrime Detection

Kamal Kant — 2026-09-30

The rapid expansion of Internet-connected systems has increased the volume and diversity of malicious network activity, creating a continuing need for intrusion detection methods that can recognize attacks from network traffic patterns rather than relying only on previously known signatures. This article presents a Hybrid CNN–Transformer (HCNT) architecture for real-time, multi-class cybercrime detection in network traffic. The approach combines one-dimensional convolutional neural networks, which are used to learn local feature interactions, with Transformer encoder layers that model longer-range relationships through multi-head self-attention. The proposed architecture uses three dilated residual convolutional blocks, learnable positional encodings, and a two-layer Transformer encoder followed by a multi-layer classification head. Five traffic classes are considered: normal traffic, Denial-of-Service (DoS), probing, Remote-to-Local (R2L), and User-to-Root (U2R). The study uses 5,000 labelled network connection records with 41 features. Class imbalance is addressed using SMOTE, while feature standardisation is fitted only on the training portion to reduce the risk of data leakage. In the reported experiment, the model reaches 97.20% test accuracy, a macro-averaged F1-score of 0.9639, and a macro ROC-AUC of 0.9971. An INT8 TensorFlow Lite version reduces the model size and records a mean inference latency of 1.87 ms in the reported benchmark. SHAP DeepExplainer is also used to examine feature contributions for different attack classes. The results indicate that combining local feature extraction with global dependency modelling can provide an effective framework for the stated experimental setting, while the limited dataset and absence of adversarial and long-term drift evaluation remain important limitations.