Yesterday, OpenAI, Anthropic, and Block announced the creation of the Agentic AI Foundation, joined by Google, Microsoft, AWS, Bloomberg, and Cloudflare.

The Wikimedia Movement is uniquely positioned to join this effort, representing the interests of the broad knowledge commons. Let me explain why this is a crucial opportunity.

The Agentic AI Foundation is meant to host and govern three open protocols and frameworks for AI development. Most interesting among them is the Model Context Protocol (MCP), which creates a standardized way for AI agents to connect and interact with web content, data and services.

Open protocols are generally great: they ensure interoperability, standardization, and collaborative governance, which levels the playing field and prevents single-actor domination. The MCP, hosted by the AAIF, is such an open standard.

But there is a paradox to this openness. MCP is a tool designed by Anthropic and embraced by other companies, but meant to be used by sources of data and content. It creates a standardized way of serving information to the AI agents. So the open standard is meant to help the agentic tools flourish by reducing barriers to their deployment and by creating an environment in which they can more easily operate. OpenAI offers the following explanation in its press release:

“​​As more agents begin handling real responsibility, the cost of fragmentation increases. Without common conventions and neutral governance, agent development risks diverging into incompatible silos that limit portability, safety, and progress”.

It’s important to understand that MCP underpins a new, emerging online ecosystem. In it, data and information flows are intermediated by AI agents rather than browsers and search engines. An ecosystem for machines, rather than for humans. This is sometimes called the “agentic web”. But the difference between the new ecosystem and the web is so big that it’s helpful to give this ecosystem a new name.

Ilario Valdelli from Wikimedia Switzerland introduced me to the idea of a “new information loop”: an ecosystem, in which AI agents ingest, transform and generate information, interacting with sources (like WIkipedia) not through interfaces designed for humans, but through machine-to-machine protocols. So there is a new Loop emerging, on top of the Web. The rules for how this Loop operates are being determined right now, and the MCP is one of the key elements.

So then, coming back to Wikipedia. It’s the darling of the original web, a peer produced platform that survived when countless dotcoms failed, and kept its civic, non-profit, commons-based status as the web platformized. It’s the “last best place on the internet”. In recent years, it also proved to play a foundational role in the development of generative AI technologies. This track record and indispensability give the Wikimedia Movement a clear place at the table when governance of the new Loop is discussed.

**It is the spirit and principles of Wikipedia that need to be introduced into MCP’s governance process. **The AAIF, according to Anthropic’s press release, is meant to ensure that “agentic AI evolves transparently, collaboratively, and in the public interest”. Yet currently, the members include five Big AI companies, a fintech company, Cloudflare with its vision of the web as a marketplace that it can control, and Bloomberg, representing commercial publishers.

This new ecosystem, where the web intertwines with new agentic information loops, will be unlike the World Wide Web. There is no origin story similar to that of Tim Berners-Lee, a civic-minded inventor, supported by a big public research institution. The new Loop is at its inception platformized and hyperscaled, with advertising coming just around the corner.

But we also have, unlike in the 1990s, the experiences of the open web, and a giant, peer-produced source of common knowledge that apparently is indispensable for AI development. The Wikimedia Foundation is right that “Wikipedia has never been more valuable than in the AI age”. **It’s then crucial to ensure that Wikipedia’s public value, built within the open web ecosystem, is not lost as the new agentic loop expands. **The Wikimedia Movement – Wikimedia Foundation, other Wikimedia organizations and volunteers – should take the responsibility for representing the commons in governance debates for the new loop.

With WIkimedia Enterprise, Wikimedia has swiftly responded to the challenge of generative AI. This project, which created a new Wikipedia API for commercial entities and a voluntary monetization scheme for these companies, is in my opinion one of the most important innovations in openness in the last decades. It can be seen as a way of shaping the new information loop in a way that gives Wikimedia some control over machine uses of its commons.

Still, more needs to be done. The Wikimedia Foundation, which stewards Wikipedia’s infrastructure, is focusing this year on developing technical solutions to curb bot activity, which by now seems to be out of control, and an increasing burden on the infrastructure. Recently, the Foundation declared that new forms of attribution are needed. It is the type of a governance issue that can be brought to the AAIF and its governance process.

There’s a high risk that the new information ecosystem will not be a fair one, if it is optimized only for frictionless activity of AI agents. AAIF governance debates on the agentic AI ecosystem must prioritize sustainability and fairness. If the Agentic AI Foundation truly intends to evolve ‘transparently, collaboratively, and in the public interest,’ then the voice of the digital commons—represented by the Wikimedia Movement—is not optional, but essential.