OSI launches fellowship to define and study open source AI

The Open Source Initiative (OSI) has launched a new fellowship program aimed at advancing research and policy discussions around open source artificial intelligence as governments, developers and technology companies debate what qualifies as open source AI.

The announcement was made during the United Nations Open Source Week in New York, where OSI introduced the Open Source AI Fellowship, a two-year initiative focused on studying AI models, measuring industry trends and building consensus around the organization's Open Source AI Definition (OSAID).

The fellowship is being established in partnership with Duke University Sanford School of Public Policy and its Science & Society Program. The initiative is supported by sponsors including Red Hat, Amazon Web Services (AWS), Google, Automattic and Mozilla.

OSI named Gabriel Toscano as the inaugural fellow. Toscano recently completed a master's degree in public policy with a focus on technology policy at Duke University.

His research during the first year will focus on three areas: analyzing how value is created and distributed within open source AI ecosystems, evaluating AI models released since the publication of the Open Source AI Definition, and organizing discussions among industry stakeholders to develop common interpretations of what constitutes open source AI.

Among the models expected to be studied are recent releases such as the Gemma 4 family, which have entered the market amid ongoing debate over openness, licensing, transparency and access to training data.

OSI Executive Director Duane O'Brien said the fellowship is intended to help establish a clearer understanding of open source AI as adoption accelerates.

"Open Source works because we share a common understanding of what it means," O'Brien said. "As AI systems become more prevalent, that shared understanding matters more than ever."

The initiative comes at a time when policymakers and industry groups are increasingly scrutinizing claims that AI models are "open source." While many developers publish model weights or portions of their code, questions remain about access to training data, transparency of development processes and the rights granted to users.

OSI has sought to address those issues through its Open Source AI Definition, which outlines criteria for determining whether an AI system can legitimately be described as open source.

The fellowship's research agenda will also explore broader questions, including how open source AI contributes to trust and accountability, how communities improve AI systems over time, and how governance frameworks may affect open source development.

As part of the program, Toscano will lead a series of multi-stakeholder discussions bringing together developers, researchers, policymakers and industry participants. The first gathering is scheduled for October in Raleigh, North Carolina, alongside the All Things Open conference.

The fellowship will also contribute research to the Open Technology Research Symposium in Barcelona and support the development of the Open Technology Research Network, a collaborative initiative involving OSI, OpenForum Europe, the Open Knowledge Foundation and the Digital Infrastructure Insights Fund.

For enterprises and governments evaluating AI adoption strategies, the program reflects growing efforts to create common standards around openness, transparency and governance as AI becomes a critical part of digital infrastructure.

The fellowship's findings are expected to inform future policy discussions on AI regulation and help shape industry understanding of what qualifies as open source AI in an increasingly complex technology landscape.