POSE: Phase I: OpenTrustLLM: An Open-Source Ecosystem for Trustworthiness in Large Language Models

This Pathways to Enable Open-Source Ecosystems (POSE) project involves the creation of an open, community-driven framework to evaluate and enhance the trustworthiness of large language models (LLMs). Large language models are increasingly utilized in sectors such as healthcare, finance, education, and national security, yet concerns about their reliability, safety, and transparency remain significant. This project establishes a collaborative ecosystem that enables stakeholders to assess trustworthiness using open standards and transparent processes. By promoting confidence in artificial intelligence technologies, this project advances national health, economic growth, and benefits all Americans. The effort contributes to scientific and technological understanding by promoting rigorous evaluation practices and facilitating education around trustworthy artificial intelligence development. The project strengthens United States leadership in artificial intelligence safety and reliability, benefiting academic researchers, industry professionals, government agencies, and the broader public through more dependable artificial intelligence applications. This Pathways to Enable Open-Source Ecosystems (POSE) project revises an existing trustworthiness evaluation platform into a sustainable open-source ecosystem named OpenTrustLLM. The project addresses the challenge of fragmented trustworthiness evaluation methods by constructing a unified infrastructure with distributed community governance. Key objectives include refactoring the current framework for modular contributions, developing continuous integration workflows, establishing a long-term governing committee, and expanding an engaged user and developer community. The project integrates multiple evaluation tools to cover critical trustworthiness dimensions such as robustness, privacy, and safety. Technical approaches include open-source software engineering best practices, black-box evaluation protocols applicable to both proprietary and open-source large language models, and proactive community education. The outcome is a scalable and sustainable ecosystem that enables systematic trustworthiness assessment for a wide range of large language models deployed in real-world applications. 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.