Abstract: With the advancement of artificial intelligence, autonomous unmanned systems with embodied intelligence are increasingly deployed in open environments to perform complex tasks. Due to their limitations in perception and decision-making, a single unmanned system struggles to efficiently adapt to various environments, which degrades the capability to address challenges in communication, networking, and message exchanging. This article studied the state of arts in terms of open message exchanging of unmanned systems, and proposed a dependable interaction framework, including zero-trust authentication and cross-domain trust evaluation, unknown entities and relationships identification with open environment modeling based on knowledge graphs, retrieval-augmented hallucination detection based on generalization bound modeling of LLMs. Finally, three application cases demonstrate the proposed technical paths. This article provides a systematic approach to achieve trustworthy, dependable interaction of unmanned systems in open environments.