The emergence of Agentic Artificial Intelligence (AI) marks a significant shift from passive generative systemsto autonomous agents capable of reasoning, planning, and executing actions with minimal human intervention.These agents increasingly interact with external tools, APIs, cloud resources, and software environments, enablingadvanced automation across domains such as software engineering, cybersecurity, business operations, andscientific research. However, the ability of AI agents to generate and execute code autonomously introducessubstantial security challenges, including unauthorized resource access, privilege escalation, prompt injectionattacks, malicious code execution, data leakage, and supply chain vulnerabilities. Traditional security mechanismsdesigned for human-operated applications are often insufficient to address the dynamic and autonomous nature ofagent-driven execution environments.This study proposes a distributed and sandboxed execution architecture for securing autonomous AI agentsthrough the integration of WebAssembly (Wasm), container runtimes, and Kubernetes-based orchestration. Theproposed framework adopts a defense-in-depth approach that isolates AI-generated actions within lightweightWasm sandboxes while leveraging Kubernetes for scalable workload management, policy enforcement, resourcegovernance, and runtime monitoring. By combining cloud-native technologies with secure execution principles,the architecture aims to minimize attack surfaces, contain potentially harmful agent behaviors, and provideauditable execution pathways for autonomous operations.A design science research methodology is employed to develop and evaluate the conceptual framework. Thearchitecture is analyzed against common threat scenarios associated with Agentic AI, including code injection,unauthorized system interactions, and compromised execution modules. The findings indicate that WebAssemblybased sandboxing offers stronger isolation and reduced overhead compared to traditional virtualizedenvironments, while Kubernetes enhances scalability and operational resilience. The study contributes a vendorneutral security model for autonomous AI systems and provides practical guidance for organizations seeking todeploy trustworthy, secure, and scalable Agentic AI infrastructures. Future research directions includeconfidential computing integration, adaptive policy engines, and decentralized security frameworks for multiagent ecosystems.