A Stanford research scholar asks whether the multi-stakeholder model can guide AI. The record of more than a dozen IRPs offers answers he hasn’t considered.

Charles Mok’s recent CircleID article, “Do We Need Alignment Between Internet Governance and AI Governance?”, raises important questions about whether the multi-stakeholder model that governs the Internet can—or should—guide the emerging architecture of AI governance.

But his analysis contains a striking omission. And that omission is not minor. It is the very evidence that would answer his questions.

Mok writes that he has “never heard” the view that “multi-stakeholder” masks corporate or state capture. He questions whether ICANN’s accountability mechanisms are “performative,” and says he does not agree with that characterization.

With respect, the record suggests otherwise. And that record is public.

What the IRP Docket Shows

Since the 2012 new gTLD round, more than a dozen Independent Review Processes have been filed against ICANN. Filing an IRP costs millions of dollars in legal fees—so people did not file because they had nothing better to do. They filed because they were aggrieved. Because the system had failed them.

Some cases resulted in published determinations. Others were settled behind closed doors. Several produced evidence of institutional failure that was never adjudicated on its merits.

• Take .africa – my case. We won an IRP, proving the ICANN Board violated its own bylaws. Discovery revealed staff misconduct, panel corruption, and a CEO quietly rewriting endorsements. That evidence never saw a trial. A procedural doctrine called Judicial Estoppel blocked us—not because our facts were wrong, but because our Lawyers had truthfully argued, during the IRP, that we had no other forum.
• Take .amazon – which built directly on the precedent we established. A years-long fight involving a multi-country dispute, resolved only after political intervention.
• Take .persiangulf – Take .persiangulf—filed the same year as .africa, cited our precedent. The case dragged on for years. In January 2026, it was finally dismissed—not because the claims lacked merit, but because the claimant company had been legally dissolved in Bahrain since 2018. For eight years, no one—not ICANN, not the panel, not the lawyers—noticed that the claimant had no legal standing. Eleven years, over $274,000 in panel fees, and the case ended not with a ruling, but with a corpse. That is not a failure of one party. That is a system-wide collapse of due diligence.
• Take .registry – won its IRP, was named prevailing party, and ICANN was ordered to pay $235,000 in fees. The applications for .INC, .LLC, and .LLP? Still denied. Winning meant nothing.
• Take Namecheap—where the claimant prevailed on transparency claims, yet the contracts at issue were maintained without change.

The common thread is not failure to win. It is a failure of the system to provide a meaningful remedy even when claimants prevail.

The full docket is public. Anyone can review it: ICANN.org/irp

Why This Silence Matters

Mok asks whether “civil society was a decorative layer” in Internet governance. In case after case, it was not decorative—it was excluded, despite winning on merit. That is not a theory. Those are the docket numbers.

He asks whether the multi-stakeholder model masks capture. The IRP record answers: yes, it can—and the documentation exists to prove it.

He asks how to prevent “digital colonialism 2.0” in AI governance. The .africa case offers a warning:

The African Union was advised by ICANN to join ICANN as a member and then oppose DCA’s .africa—not on technical grounds, but through the Governmental Advisory Committee. An African body, turned against an African applicant, using global governance procedures. That is not multi-stakeholderism. That is sovereignty subcontracted.

That is the documented truth.

That is the architecture of exclusion. And it is not accidental. It is structural.

ICANN’s Limited Mandate – And Why AI Needs Something New

Mok holds up ICANN as a potential model for AI governance. But this comparison ignores a fundamental truth: ICANN’s mandate was never designed to govern content, data, cybersecurity, or digital markets.

ICANN’s mission, as defined in its Bylaws and reaffirmed in the Affirmation of Commitments, is narrow and technical: coordinate the global DNS, ensure its stable and secure operation, and maintain a unique authoritative root. When the Internet became commercial in the 1990s and content—data, commerce, communication—began flowing through the pipes ICANN manages, the organization wisely did not extend its mandate to govern that content. It stayed at the tool level.

That was the right decision. ICANN coordinates the address book. It does not—and should not—tell you what to write in the letter.

Yet now we are told that this same institution, with its narrow technical remit and its documented accountability deficits, should serve as a template for governing artificial intelligence—a technology that shapes labor markets, electoral integrity, military strategy, and the very boundaries of human cognition.

The scholar Chen Jian has written that digital colonialism in the AI era operates through “monopoly control of digital technology and data resources” to construct a “digital colonial system subservient to Western powers.” If AI governance simply replicates ICANN’s structure—with its unaccountable staff, its procedural traps, its gap between winning and remedy—it will not prevent this colonialism. It will encode it.

AI governance does not need a photocopy of ICANN. It needs an honest institution—one with no baggage of policy abuse, no decades of documented failures, no IRP docket full of claimants who won but walked away empty-handed.

There are capable institutions that could take up this mantle. Or new ones could be built. But the starting point must be different: a mandate that reaches conduct, not just board decisions. A forum where evidence is heard, not blocked. A structure where the Global South has a voice—not as a decorative layer, but as co-author of the rules.

The question Mok should have asked is not whether ICANN can be a model for AI governance. It is: who will build the institution that can actually govern, without replicating the immunities of the past?

The Geopolitical Context Mok Avoids

Mok writes from Stanford, citing ICANN as a potential model for AI governance. But one must ask: why this framing, and why now?

When India hosted the AI Impact Summit in February 2026 and secured 88 signatories—including the US, China, Russia, and the EU—for the New Delhi Declaration on AI governance, it signaled something significant: developing nations want a seat at the table. They want AI governance that serves the many, not just the few.

The declaration’s emphasis on democratizing AI resources, affordable connectivity, and locally relevant innovation stands in stark contrast to the “innovation first, regulations never” posture that has defined certain approaches.

And when the US delegation explicitly rejected “global governance of AI” at the same summit, the message was clear:some powers want to keep control centralized.

Mok’s framing—holding up ICANN as a model while dismissing critiques of it—fits a pattern. It suggests that Western-designed institutions, with all their flaws, should set the template. It ignores that the Global South has its own experiences, its own grievances, and its own vision for governance.

What AI Governance Must Build Instead

Mok is right that AI governance cannot simply copy Internet governance. The starting points are different. The stakes are different. The power asymmetries are deeper.

But if AI governance looks to ICANN for lessons, it must look with open eyes—at the docket, at the cases, at the gap between winning and being heard. That gap is not a bug. It is the design.

What AI governance needs is not a replica of ICANN. It needs:

1. Reachability of conduct. A forum that cannot examine misconduct beyond boardrooms is not a forum—it’s a filter.
2. Anticipation of procedural traps. Doctrines like Judicial Estoppel can be weaponized to foreclose scrutiny. Future models must build safeguards.
3. The record as remedy. When courts close, history remains. Documented evidence of misconduct must not disappear because a procedural door shuts.
4. Standing for the excluded. The Global South must have a voice—not as a “decorative layer,” but as co-author of the rules.

A Closing Note to Charles Mok

Charles, your article asks the right questions. But the answers lie in places you have not looked—in the IRP docket, in the testimony of those who tested the system and were blocked from being heard, in the evidence that never saw trial.

The view that multi-stakeholderism masks capture is not new. It is not a fringe opinion. It is documented. It is the docket numbers. It is the .africa case, the .amazon case, the .persiangulf case, and others.

I do not offer this as a grievance. I offer it as documentation. If AI governance is to avoid the same traps, it must begin not with theory, but with the record of what actually happened when people tried.

That record exists. It is public. And it tells a story your article left out.