NSF Safe-OSE: Center for Security-Assuring AI Agent Communication and Interoperability Standard and the Open-Source Ecosystem

Artificial intelligence (AI) agents are increasingly used in government services, business systems, research environments, and other digital infrastructure, yet they currently lack a safe and reliable way to communicate across different organizations and technologies. This project strengthens the Safety for Open‑Source Ecosystems (Safe‑OSE) mission by improving the safety, security, and trustworthiness of the Natural Language Interaction Protocol (NLIP), an emerging open standard that enables AI agents to exchange structured messages. By enhancing the reliability of this communication layer, the project supports the national interest in secure and resilient digital infrastructure and reduces risks of unintended actions, data exposure, or system failures. The project also expands participation in open‑source development by producing freely available tools and resources that help developers, educators, and organizations adopt safer AI systems. These improvements benefit society by making AI‑enabled services more dependable and easier to secure. The project will develop and evaluate security mechanisms that operate directly at the NLIP protocol layer, using NLIP’s standardized message envelopes and control fields as enforcement points for validation, authorization, and policy constraints. The team will analyze vulnerabilities in agent‑to‑agent and agent‑to‑user communication—such as authorization weaknesses, session hijacking, cost‑amplification misuse, memory poisoning, prompt injection, and semantic manipulation—and design protocol‑level mitigations that are generalizable across frameworks. Methods include implementing open‑source security extensions, building validation and conformance tools, and testing protections across multiple real‑world agent frameworks and enterprise systems. This unified approach enables consistent, cross‑platform safeguards that are difficult to achieve when defenses are implemented separately within individual applications or vendors. The resulting tools, standards‑aligned mechanisms, and evaluation results will advance the science of securing AI‑mediated communication and strengthen the broader NLIP open‑source ecosystem. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.