Why Mark Zuckerberg wants to redefine open source so badly

LONDON -- Integrating artificial intelligence into the world of open source is no simple task. While it's true that AI has deep roots in open source, with notable examples like IBM's Granite Large Language Models (LLM) and RHEL AI fully embracing open-source principles, the reality is that many high-profile AI models, like Meta's Llama, fall short of being truly open source. Despite Meta CEO Mark Zuckerberg's assertions to the contrary, a panel of open-source experts at the State of Open 2025 conference debunked this claim.

There's now an official definition for open-source AI, but the debate rages on.

Emily Omier, a respected consultant for open-source start-ups, stressed that open source is not a sliding scale but a clear-cut standard defined by the Open Source Initiative (OSI). "You're either open source, or you're not. If you have an OSI-approved license, you're open source. Without it, you're dealing with a different kind of license altogether," she explained.

Meta's Llama doesn't meet this standard because it withholds essential elements like training data and methodology, and it restricts transparency and community contributions.

Moreover, Llama imposes licensing restrictions on its users. For instance, if you create a highly successful AI program using Llama's code, you'll owe Meta royalties. That's not open source by any measure.

Stefano Maffulli, the executive director of the OSI, shared his thoughts with me: "We align with everything Zuckerberg has stated in theory. If only Meta would lift the restrictions on their license, we'd be on the same page. As it stands, Llama poses a risk to developers; it's too opaque to be safely used and comes with a license that keeps Meta in control of any innovations."

Maffulli went on to say that Zuckerberg was "pressuring the industry to follow his lead."

It's widely accepted that open source is the optimal approach for software development. Zuckerberg himself acknowledges this, stating, "We've thrived on the innovations from the open-source ecosystem by making leading tools like PyTorch, React, and many others freely available. This strategy has proven effective for us when we stick to it over time."

However, all those tools are released under OSI-approved licenses. When it comes to AI, Zuckerberg seems intent on redefining open source. Why? Peter Zaitsev, co-founder of Percona and numerous open-source startups, suggests that "Zuckerberg is muddying the term to mislead regulators."

This is crucial for Meta's plans in the European Union (EU), where the newly enacted EU AI Act offers an exemption for AI systems distributed under free and open-source licenses, potentially saving Meta billions.

Meta is therefore pushing to redefine "open source" in the context of AI, all while keeping critical components under wraps. According to Meta, "The current open-source definitions for software don't fully address the complexities of modern AI models. We're dedicated to collaborating with the industry to develop new definitions that ensure safety and responsibility within the AI community."

In essence, Meta wants to redefine open source in a way that benefits them exclusively.

Yet, this goes against the very essence of open source, regardless of Meta's arguments.