Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- Stefano Maffulli, the OSI’s first executive director, is leaving.
- His biggest challenge was working on AI and open source software.
- Under his leadership, the OSI created the Open-Source AI Definition.
The Open Source Initiative (OSI) is facing a major leadership transition as it launches the search for a new executive director.
Stefano Maffulli, the group’s first executive director, is set to step down in October to pursue work in open-source AI and data governance. Under Maffulli’s leadership since 2021, OSI moved from a volunteer-centric group to a globally recognized nonprofit, notably releasing the Open Source AI Definition (OSAID) 1.0, which, like the name suggests, established a standard for open-source AI licensing.
The OSI, founded in February 1998, successfully advocated for open-source principles and stewarded the Open Source Definition (OSD). For the last few years, however, the organization has grown slowly, even though new open-source challenges have emerged. When Maffulli took the position as the organization’s first executive director, he faced the rise of difficult issues.
Also: Linus Torvalds warns Linux devs: Stop cluttering patches with automated, useless links
As Maffulli said when he was appointed, “Open source software is everywhere, but its definition is constantly being challenged. The zombies of shared source, limited-use, and proprietary software are emerging from the graves.”
When he took the job in 2021, he recalled, “Some people thought OSI’s work was done. The Open Source Definition had ‘won,’ embraced even by organizations that once resisted it.” He went on to say they thought, “there was nothing left to do. I didn’t see it that way.”
He was right.
Open source challenges in the 2020s
For example, as Maffulli pointed out, the launch of the Apple Store for the iPhone in 2008 made it impossible for users to modify and run open-source applications on the smartphone. In October 2021, GitHub announced Copilot. Maffulli said it posed an even bigger challenge: “In a world where systems aren’t traditionally ‘programmed,’ what does ‘source code’ mean?”
It would be this last issue, the role of AI in open source, that would bedevil the OSI for years.
Maffulli’s tenure culminated in the release of OSAID 1.0, a widely endorsed framework defining what constitutes an “Open Source AI.” According to the official OSI definition, open-source AI systems must grant users the freedom to use, study, modify, and share the system — including models, weights, and code — for any purpose. However, pragmatists influenced the definition to allow for sufficiently detailed information about the data used for training, instead of requiring complete open access to datasets. This compromise balances transparency against legal and ethical concerns, such as proprietary or sensitive data.
Also: AI is creeping into the Linux kernel – and official policy is needed ASAP
Leading open source figures and foundations, including Mozilla, SUSE, Bloomberg Engineering, and the Cloud Native Computing Foundation, have endorsed the OSAID. Critics and supporters alike acknowledge its importance as a foundation for AI licensing, transparency, and code accessibility. That said, the OSI freely admits that the OSAID is “a work in progress.” This view is echoed across academic, industry, and policy circles. The consensus is that the definition represents a significant, if imperfect, advance for open-source principles in the complex world of AI.
Indeed, many open-source developers and leaders, such as Richard Fontana, Red Hat’s principal commercial counsel, disagree with this definition. They want the data and weights to be more open than they are under the OSAID 1.0. As Luca Antiga, CTO at AI agent company Lightning, and a top contributor to the vital AI library PyTorch, wrote in The New Stack, by not incorporating weights into the OSAID, “The OSI is leaving a gaping hole that will make licenses less effective in determining whether OSI-licensed AI systems can be adopted in real-world contexts.”
Where OSAID goes from here
They’re not alone with such concerns. Peter Zaitsev, CEO of the open-source data company Percona, believes “OSI continues with its flawed ‘one size fits all’ approach rather than helping to better define the ‘spectrum’ for open source and AI.” In addition, when “training data is not freely available for everyone, it is not the same as ‘open source.'” This debate shows every sign of continuing. The new OSI Executive Director will need to deal with these issues.
What both supporters and opponents of the OSAID can agree on is that many AI companies, such as Meta with Llama, X with Grok, and Microsoft’s Phi-2, claim to be open source, but they’re not. They want to redefine open-source AI for their own profits while restricting anyone else from commercially using their code, weights, and data.
Also: You can get Nvidia’s CUDA on three popular enterprise Linux distros now – why it matters
OSI’s board is committed to continuing its emphasis on licensing, policy, and building the future of open source AI. Deborah Bryant has been named interim executive director during the transition. The organization credits Maffulli for modernizing its operations, expanding international partnerships, and strengthening OSI’s voice in global technology policy.
As Tracy Hinds, chair of the OSI board of directors, said in a statement, “Stefano helped build a strong and resilient foundation for OSI during a time of extraordinary change in the tech world. We are grateful for his dedication to OSI’s mission and to the global open source community. The board is committed to maintaining our course and continuing the important work underway — especially around licensing, policy, and the future of open source AI.”
Want to follow my work? Add ZDNET as a trusted source on Google.