A Unified Architecture for Trustworthy Autonomous Intelligence: Integrating Agentic AI, Edge Computing, Digital Twins, Cybersecurity, and Explainable Decision-Making

The convergence of agentic artificial intelligence, edge computing, digital twins, cybersecurity, and explainable AI (XAI) is producing autonomous systems capable of multi-step reasoning and real-world action, yet governance and security frameworks have not kept pace with this shift. Recent surveys document that agentic systems introduce reasoning and coordination capabilities absent from earlier reactive AI, while simultaneously expanding the attack surface through prompt injection, memory poisoning, and inter-agent collusion. Parallel literatures on trustworthy edge intelligence and AI-integrated digital twins emphasize simulation-based verification and resource-aware trust management, and separate governance literatures (NIST AI RMF, the EU AI Act, ISO/IEC 42001) note that existing instruments were designed for single-step predictive systems rather than autonomous multi-step action. This paper synthesizes these largely separate literatures into a proposed six-layer architecture: edge sensing, digital twin verification, agentic reasoning, explainable decision-making, governance and compliance, and human oversight intended to close the gap between autonomous capability and verifiable trust. It maps a cross-layer threat taxonomy drawn from documented attack classes, positions explainability as a mandatory gate rather than an optional add-on, and evaluates the coverage of current governance instruments against each layer, identifying agentic reasoning as the area of greatest regulatory and technical gap. A proposed (not executed) evaluation protocol is outlined for future empirical validation. The paper's contribution is conceptual and integrative: it does not report new experimental results but provides a structured, literature-grounded foundation and research agenda for engineering and governing trustworthy autonomous systems.