We are building HG SafeExec, an autonomous AI agent system for secure task execution, sandboxed code operations, automated testing, and verifiable audit trails. Our architecture combines agent orchestration, isolated execution environments, policy-based permissions, failure recovery, and human oversight for high-risk actions. We aim to demonstrate how AI agents can perform real-world work reliably and safely.