The Open Source Initiative has launched a two-year research fellowship to combat “openwashing” and officially define standard criteria for open-source AI models.
The Open Source Initiative (OSI) has announced the creation of the Open Source AI Fellowship during the United Nations Open Source Week in New York. Developed alongside Duke University’s Sanford School of Public Policy, the two-year program is designed to conduct original, peer-reviewed research and establish an international consensus on what technical requirements a model must meet to legally carry an open-source designation.
The initiative launches amid severe geopolitical and corporate friction over “openwashing,” where dominant tech firms label models as open source despite locking away underlying training datasets or enforcing restrictive user licensing agreements. Operating with foundational backing from industry pillars—including Google, AWS, Red Hat, Mozilla, and Automattic—the fellowship intends to establish a neutral baseline for AI openness to inform looming global regulatory frameworks, such as the European Union’s newly minted Tech Sovereignty package.
Gabriel Toscano, a technology policy researcher from Duke University, will serve as the inaugural fellow. Toscano’s initial agenda centers on analyzing recent model architectures—specifically evaluating new releases like Google’s Gemma 4 family—against version 1.0 of the official Open Source AI Definition (OSAID). His research will systematically measure value flows in public ecosystems, tracking how architectural community modifications iterate and cycle back into foundational AI frameworks.
The fellowship will anchor its stakeholder outreach later this year with an introductory assembly in Raleigh, North Carolina, followed by a formal presentation of data at the Open Technology Research Symposium in Barcelona.
