Interoperability Is a Means, Not a Finish Line for AI Sovereignty

The Tech Policy Press piece — “Why AI Sovereignty Depends on Interoperability Standards” — makes a clean, technical case: if states want control over AI, they need shared standards so systems can talk to each other. Sounds right as far as it goes. But sensible isn’t sufficient. Here’s what nobody tells you: standards are power instruments dressed up as plumbing diagrams.

Standards as power, not neutral plumbing

The article’s strongest move is on the basics: shared formats and APIs make it easier for governments to swap vendors, avoid lock-in, and audit models. That’s the technical face of sovereignty — optionality.

But it quietly treats standards as neutral public goods. They’re not. They’re written by specific institutions, with specific interests, trying to freeze their view of “normal” into code and compliance.

If OpenAI, Google, or Microsoft dominate standard-setting bodies — as they’re likely to, given their control of talent and infrastructure — then “interoperability” risks becoming alignment with their architectures and incentives. Yes, the EU can pass strict data rules, and China can demand localization. That’s table stakes. If the standards underneath those policies quietly favor closed model formats or proprietary logging hooks, then states have swapped old dependencies for new ones.

You don’t get sovereignty from a menu of similar suppliers who all share the same technical assumptions and telemetry channels.

As someone who used to run operations in a large enterprise, I’ve seen this movie: the “open” interface that turns out to have undocumented behaviors, proprietary throttling rules, and support paths that only the original vendor can navigate. On paper, you’re portable. In practice, you’re stuck.

Standards only serve sovereignty if their governance pulls in small states, independent researchers, and civil society — not just hyperscalers and a few wealthy capitals.

Who writes the rulebook matters more than the rulebook

The Tech Policy Press article nods at geopolitics but doesn’t really stay there. That’s the real question: whose preferences are being frozen into the standards that everyone else has to live with?

Look at telecom. The 3GPP process produced global wireless standards that did, in fact, power massive growth. It also cemented the influence of the firms and countries that could afford to carpet-bomb the committees with engineers and patent lawyers. Those with R&D capacity shaped the stack; others became takers, not makers.

AI standards will follow the same pattern if we’re not careful. If Western firms or Chinese state-backed players dominate the working groups, then “best practice” will tilt toward their commercial and political needs. Compliance then becomes a specialized certification industry run by the same giants that define the rules.

Two outcomes follow from that. First, conformity isn’t just a technical cost — it’s a recurring operational expense, with audits and updates sold by the rule-makers. Second, smaller or poorer countries will adopt the prevailing standard not because it preserves autonomy, but because it’s the cheapest, most supported path.

If your regulator must rely on certifications issued by entities headquartered elsewhere, what you have is managed dependence with nicer branding.

Innovation, drag, and who gets to move fast

Critics of standards like to say they slow innovation and turn exploration into committee meetings. Spare me. The real issue is: whose innovation gets slowed, and who gets to sprint ahead inside the rules they wrote?

Standards undeniably close off some paths. That’s often the point. You trade a bit of local chaos for lower integration costs, shared security baselines, and fewer nasty surprises in critical infrastructure. The trick is making sure you’re slowing the right things: runaway network effects and opaque interfaces, not genuine experimentation.

You can design standards to be modular and versioned, with clear upgrade paths. You can hardwire transparency requirements — logging, audit hooks, provenance metadata — into conformance tests. That setup penalizes fast-and-loose proprietary advantage, but it opens space for smaller providers that can adopt the standard without reverse-engineering a walled garden.

We’ve seen a version of this before with the web. Open standards like HTTP and HTML didn’t kill innovation; they stopped any single browser vendor from unilaterally deciding what “the internet” was. Then came app stores — vertically controlled ecosystems with their own de facto standards — and suddenly a tiny number of companies set the rules for distribution, payments, and content. AI can drift either way.

Practical governance gaps the article misses

The Tech Policy Press piece mostly stays in the technical lane and glides past enforcement and participation. That’s where things break.

Standards without accessible test suites, reference implementations, and dispute-resolution processes are just PDFs with acronyms. Small states often lack the staffed-up agencies and labs to even sit in the standards meetings, much less build independent conformance tests. Civil society groups and universities are rarely funded to do this grinding, unglamorous work.

If governments actually want standards to underpin sovereignty, they need to bankroll participation — travel, engineers, testing infrastructure — instead of just issuing policy memos about “adhering to international norms.”

And look — interoperability on its own won’t shield anyone from subtle forms of control, like alignment biases or content filtering defaults, if the standards only touch APIs and file formats. Real sovereignty in AI will require standards that expose audit interfaces, support verifiable provenance, and embed rights-preserving control points instead of just optimizing for throughput.

Otherwise, you don’t get interoperable freedom. You get interoperable gatekeeping.

Tech Policy Press is right that interoperability standards will sit at the heart of AI sovereignty debates; the open question is whether those standards become scaffolding for shared capacity or rails that quietly steer everyone into someone else’s station.