Journal of Future Internet and Hyperconnectivity (e-ISSN: 3048-9210)

Cloud Sentinel-AI: An Explainable Behavioral-Provenance and Self-Healing Security Framework for Autonomous AI Agents in Multi-Cloud Environments

2026-10-31T05:26:14+00:00

The rapid adoption of autonomous artificial intelligence (AI) agents in cloud environments has introduced new security challenges due to their ability to dynamically select tools, access heterogeneous resources, interpret external information, and modify execution plans at runtime. Conventional authentication and static authorization mechanisms are often insufficient to capture the evolving security context of such agent-driven activities. This study proposes CloudSentinel-AI, an explainable, behavior-aware, and self-healing security framework for protecting autonomous AI agents operating across multi-cloud environments. The proposed framework continuously constructs a behavioral provenance model that links agent identity, task context, data sources, tool invocations, API operations, cloud resources, and resulting actions. A hybrid detection mechanism combines behavioral analysis, contextual risk assessment, and provenance-deviation analysis to identify potentially malicious or unauthorized agent behavior. An explainable decision layer associates detected risks with supporting evidence and determines proportional security responses, including additional verification, capability restriction, credential revocation, and agent isolation. The proposed prototype architecture integrates Python, FastAPI, PostgreSQL-compatible storage, graph-oriented provenance representation, Docker-based deployment, a web-based monitoring dashboard, and a Flutter mobile client. To support reproducible evaluation, the CloudSentinel-AgentSec Dataset (CASD) is proposed as a synthetic benchmark containing normal and adversarial autonomous-agent sessions. The evaluation methodology considers precision, recall, F1-score, false-positive rate, detection latency, containment latency, and system resource overhead. No performance claims are made until the prototype is evaluated using the defined benchmark. The proposed work therefore provides a reproducible technical architecture and systematic evaluation methodology for improving the security, explainability, and resilience of autonomous AI agents in multi-cloud environments.

Policy-as-Code and Event-Driven State Reconciliation for Automated Multi-Cloud Security

2026-10-28T11:43:16+00:00

Multi-cloud workloads inherit a different identity model, resource defaults, and audit interface from each provider, so a security control is enforced unevenly, and a change made outside the deployment pipeline leaves a resource exposed until someone notices. This paper presents a framework that closes this exposure window at both ends. Security policy is written in Rego and evaluated against the Terraform plan inside the CI/CD pipeline, so a violating definition fails the build before apply. At runtime, AWS CloudTrail and Azure Activity Log events are normalised into a provider-neutral record and compared attribute by attribute against a signed, Git-backed baseline. A change made by a principal other than the pipeline deployment identity is reverted through a scoped provider API call, subject to rate limits, change-freeze windows, and a dry-run mode. The prototype was evaluated on a synthetic AWS and Azure testbed with eight scenarios: six injected drift scenarios, one approved pipeline deployment used as a false-positive control, and one insecure definition submitted for pre-deployment rejection. Of the nine system components, five were exercised by live scenarios, and four were implemented without scenario-level testing. Accuracy, precision, recall, F1, detection and remediation times, exposure window, rollback success, policy-gate effectiveness, and false-positive rate are defined in the Evaluation Metrics subsection, but measured values are not yet reported, so this abstract makes no quantitative performance claim. The evaluation is limited to a synthetic environment with two providers, and the protection is only as sound as the policy set it enforces.

Edge-Federated CNN Framework for Distributed Acne Severity Classification Across Devices

2026-10-29T08:32:48+00:00

Privacy, device constraints, and data heterogeneity affect acne severity assessment from facial photographs. This study presents a privacy-oriented edge-federated framework using MobileNetV3 for four acne-severity categories: mild, moderate, severe, and very severe. The reported dataset contains 25,000 facial images, including a 5,000-image held-out test set, with 224×224 input resolution and approximately 100 clients per round over 200 communication rounds. The reported 2.1 MB model achieved sub-second inference across four tested device classes. Federated training uses local optimization with server-side weighted aggregation, while privacy protection incorporates update noise and secure aggregation. The reported communication volume decreased from 450 GB to 100 GB (a 77.8% reduction), while the target performance was reached in 154 versus 200 communication rounds. Several experimental details require further verification, including data partitioning, preprocessing, privacy parameters, attack methodology, confusion-matrix counts, and baseline metrics. The framework is positioned as an automated decision-support approach rather than a substitute for clinical assessment.

NEXORA: An Integrated Student Productivity Workspace with AI-Based Resume Analysis

2026-10-15T07:13:59+00:00

Students often rely on multiple independent tools to manage academic resources, process documents, and prepare resumes, resulting in a fragmented workflow and requiring them to switch between different platforms. NEXORA addresses this problem by providing an integrated student productivity workspace within a single Django-based web application. The system provides a Resource Manager that allows students to organize study materials and useful resources into collections, with features such as starring, search, summaries, keywords, and a recoverable Trash. The Resume Analyzer uses Google's Gemini API to analyses uploaded resumes and generate structured feedback, including an overall score, ATS-related score, strengths, areas for improvement, and prioritized suggestions. The Document Toolkit provides eleven document-processing utilities for common operations such as merging, splitting, compressing, watermarking, rotating, page numbering, PDF creation, and conversion between PDF, DOCX, PPTX, and image formats. A Dashboard brings these modules together by providing usage statistics and a recent-activity view. The system is implemented using Django 5.2 as the backend framework, with PyMuPDF for PDF processing, python-docx and python-pptx for Office document handling, the Google Generative AI SDK for AI-assisted analysis, and Bootstrap 5 for the user interface. The application follows a modular architecture in which authentication, resource management, document processing, resume analysis, and dashboard functionality are organized into separate components while operating under a common user account. The resulting system provides students with a centralized workspace for organizing academic resources, processing documents, and improving resume content, reducing the need to switch between multiple independent applications for these activities.