This Is What Un-Governed AI Looks Like

One government. One lab. One deal.

That’s the sentence. Everything else is context.

OpenAI is in negotiations to offer the United States government a 5% equity stake in its operations — part of a broader effort to convert political opposition into alignment. The administration would become a financial stakeholder in the most prominent AI company in the world. No international process. No transparency mechanism. No other parties at the table.

This is what un-governed AI looks like.

The Inversion

The people building the most consequential technology in human history are doing what companies have always done with governments they need to manage: they are making the government a partner rather than a regulator. Once you hold equity, your incentive is for the asset to succeed. Oversight becomes complicated. Independence disappears.

This is not a conspiracy. This is basic organizational behavior. It doesn’t require bad intentions from anyone involved. It follows mechanically from the structure of the relationship: you cannot be an independent regulator of something you’re financially dependent on doing well.

The result is a governance structure where the most powerful AI system on earth answers primarily to one government, which now has financial reasons to want it to succeed, which means the two most powerful entities in that system — the lab and the government — are now aligned with each other and structurally misaligned with everyone else.

Billions of people who are not American. Every other government. Every institution. Every individual whose life this technology touches.

Nobody asked them.

Why There Is No Local Fix

The reflex response to this is regulatory: pass a law, create an agency, require disclosure. These are not wrong. They are insufficient.

A law passed by one government can be rescinded by the next. An agency can be captured, defunded, or ignored. Disclosure requirements tell you what happened after it already happened. None of these mechanisms work at the speed AI is moving, across the jurisdictions AI is operating in, at the level of accountability that actually matters: the individual action, in the moment it’s taken.

The governance problem with AI is not that we need better policy. It’s that the substrate — the systems themselves — have no accountability architecture built in. Every action taken by an AI agent on behalf of a human or institution is currently either traceable in proprietary logs that you cannot access, or not traced at all. There is no cryptographic chain of custody. There is no universal credential system. There is no kill switch that belongs to anyone other than the platform operator.

OpenAI offering the US government a stake in its operations is not a scandal. It’s a symptom. It’s what happens when the only lever anyone has over a system this powerful is financial.

What the Protocol Does

The Covenant is not a policy document. It does not ask AI companies to behave differently. It does not require governments to coordinate — though coordination would help.

It is a cryptographic architecture. A protocol. A set of rules enforced not by goodwill but by mathematics: every agent action signed and attributed, credentials held in human custody, authority mapped to existing governance structures, revocability built into the substrate rather than layered on top.

Under AIGCSEP, what OpenAI just negotiated wouldn’t change the math. A US government equity stake in OpenAI doesn’t let the government reach into the protocol and exempt American AI systems from accountability requirements. The cryptographic layer doesn’t care who owns the equity. The authentication chain doesn’t care which administration is in office.

That’s the point. That’s why a protocol matters where a policy doesn’t.

One government, one lab, one deal — that’s the problem.
One protocol, open, cryptographic, jurisdictionally neutral — that’s the alternative.

The window is still open. Not indefinitely.

Related: You Ain’t Seen Nothing Yet — the tidal wave context, and why the governance window is time-limited.

Related: Monoculture is a Governance Problem, Not a Technical One — what happens when all the systems converge on the same answers with no accountability for the convergence.

Related: Insource the Model — when there’s no protocol, vendor capture fills the gap. Why every serious institution will eventually build its own models — and what the late-90s tech insourcing wave tells us about the timeline.

Sources

• OpenAI floats giving US 5% stake to win over AI haters — Ars Technica