Biography
Sakera Begum is an Artificial Intelligence and Predictive Analytics Researcher, IT Consultant, and Data Analyst currently pursuing a Doctor of Computer Science at the University of the Potomac, USA. She holds a Master of Science in Information Technology and a Bachelor of Business Administration in Finance and Accounting.
Her professional expertise spans artificial intelligence, machine learning, cybersecurity, predictive analytics, business intelligence, software quality assurance, and data-driven decision- making. Her research focuses on AI-driven financial technologies, equitable capital allocation,
startup resilience, fraud detection, and responsible AI applications.
She has authored multiple research publications, published a book on AI-powered capital allocation, and holds a UK registered design. Her academic contributions include international conference presentations, manuscript peer reviewing for Springer Nature and Elsevier journals, and scientific committee participation.
She has received several recognitions, including a Best Paper Award, Best Researcher Award, and Enterprise Solutions Innovation Award. She is a member of IEEE, ACM, and the Analytics Society of INFORMS.
Her professional and research interests center on advancing intelligent, secure, and responsible technologies to address real-world economic, business, and technological challenges.
Abstract
Title: From AI-Powered Detection to Resilient Response: A Risk-Governed Agentic Cybersecurity Framework for Organizations
Abstract: Artificial intelligence (AI) can support cybersecurity teams by correlating large volumes of security telemetry, interpreting alerts, retrieving context, and recommending incident-response actions. Yet delegating operational authority to AI agents introduces risks that conventional detection benchmarks do not capture prompt injection through untrusted evidence, excessive tool privileges, inaccurate conclusions, poor audit ability, and unsafe containment. This paper presents RAMPART (Risk-Aware Monitoring, Policy Authorization, and Response Traceability), a conceptual, human-governed architecture for AI-assisted cyber defense applicable to resource-constrained small businesses and complex enterprises. The study synthesizes peer-reviewed agent-security research and authoritative guidance from the National Institute of Standards and Technology (NIST), the Open Worldwide Application Security Project (OWASP), and MITRE. RAMPART separates evidence collection, deterministic validation, AI-generated investigation, independent authorization, bounded action, and recovery verification. Its central design rule is that AI may propose an operational action, but model-generated text cannot grant itself permission to execute that action. A threat model, use-case analysis, and falsifiable evaluation protocol compare RAMPART with rules-only and unrestricted agent baselines across detection quality, analyst effort, response latency, security-policy violations, and recovery outcomes. No deployment or experimental results are claimed. The contribution is a structured architecture and test plan for assessing whether agentic AI improves cyber resilience without undermining accountability or introducing unacceptable operational risk.
