Middle Powers and the Shape of AI Governance: Three Experts Weigh In

Conversations about AI governance tend to revolve around the United States and China, or the Big Tech companies building the most advanced models. Middle powers usually enter the discussion as countries with lesser influence responding to decisions made elsewhere.

But these influential states have more room to shape AI than this thinking suggests. They decide which systems to buy, how they should be tested, and what standards they want to support internationally. Middle powers are already part of these decisions, and some are beginning to think more deliberately about how to use their position and respective points of leverage.

But where can middle powers really start to move the needle? To begin to answer this, New America and the Collective Intelligence Project asked three experts to weigh in. Drawing on their own work, each expert focuses on a specific part of AI governance and what middle-power governments could influence in that area there.

Marshall McLuhan famously asserted that “the medium is the message.” The communication structure enabled by technology (“the medium”) shapes society as much as the content of the communication (“the message”).

Whereas broadcasting has a one-to-many communication structure, social media’s is many-to-many. The impact of AI on many-to-many social media platforms is already visible. Platforms are flooded with AI-generated content, and AI agents may soon make up the majority of participants online. Labeling that content is unlikely to be enough to preserve a digital public deliberative sphere for humans in the long run. We may also need proof-of-personhood standards that verify a participant is human without revealing their identity.

Less visible is the new communication structure of AI chatbots and other applications. But it has just as much potential to reshape society. The structure is best characterized as many-to-one. Many human senders initiate personalized conversations with one AI model. To the user, this feels like a one-to-one conversation. Yet structurally, millions of humans talk to different copies of the same AI model. Whatever political leanings that AI model has are reproduced across millions of private exchanges. Because each exchange feels private, none of it looks like mass communication, and none of it receives the scrutiny a mass communications platform would be subject to. The result is a stronger persuasion channel than any single newspaper or broadcaster ever had—and with none of the visibility.

A democratic middle-power coalition could bring more transparency to this system. AI testing institutes could jointly benchmark and report on the persuasive capabilities of conversational AI models, and require the disclosure of political bias for widely used models. No single middle power can impose this on frontier labs, but a coalition of democracies could.

With the rapid accumulation of AI cyber incidents, the central challenge set for AI governance is increasingly one of speed and agility. Global AI governance is urgently in need of a mechanism with the capacity to respond to evolving capabilities and risks before a crisis occurs.

Middle powers have an important role to play in filling that gap, precisely because they are less likely to view global governance principally through the lens of technological and geostrategic competition. That distance gives them room to move where the two largest AI powers cannot, supporting a global approach to frontier AI governance based on predictable rules, shared safeguards, and institutional cooperation. Acting collectively, they can help identify areas where progress remains achievable even amid competition between the U.S. and China.

The clearest avenue for such cooperation is the international AI summit series. Early editions—Bletchley and Seoul—demonstrated that small groups of governments, working together with industry, can move quickly to put frontier risks on the international agenda and develop meaningful forms of cooperation. That capacity for rapid international cooperation matters increasingly as cyber, bio, and other risks emerge and evolve faster than traditional multilateral processes can typically respond.

The 2027 Geneva AI Summit, therefore, arrives at a pivotal moment in AI governance. Middle powers should encourage Switzerland to use it not simply as a forum for dialogue, but to advance tangible cooperation on cyber and biosecurity, incident reporting, evaluation, verification, and other practical safeguards. Used strategically, Geneva could allow middle powers to demonstrate that meaningful cooperation on frontier AI remains possible even in an era of great-power competition and mistrust.

What can middle powers do collectively to shape a more democratic global approach to AI governance?

The prevailing “AI race” narrative frames success as outbuilding the largest labs. A collective rethink is required. We should make freedom to inspect, switch, and repair the shared baseline for public-sector AI.

Start with open procurement terms co-written by representative panels of impacted, front-line workers and participating governments, then jointly maintained. Suppliers would give auditors access through structured transparency, make data and interfaces portable and comply with specified lawful remedies such as correction, suspension, or replacement.

Each government would state transparently who may raise a concern, who must answer, by when and what remedies are available. Public buyers could version the terms together, sharing implementation failures for every other adopter to learn from in revising the baseline.

No government would have to learn the same lesson alone. All countries great and small could shape the baseline from the outset and adopt it without financing a frontier model from scratch.

Before any tender is fixed, the same panels would review local adaptations. Governments could publish what changed, what did not and why, while naming the scope and legal basis of any withheld details.

In the public processes we facilitated in Taiwan, I learned that listening does not complete the loop: institutions must answer, and the people must be able to trace a path from harm to repair.

We can make recourse portable and keep local choices local. A race asks who wins. A coalition of like-minded middle powers asks whether the democratic world can still change course. Yes! Taiwan can help.