---
title: Privacy Is Not a Setting
author: Hilal Agil
date: 2026-06-02
source: https://hilalagil.com/essays/privacy-is-not-a-setting/
topics: Data, Privacy, AI Governance, Decentralization
---

# Privacy Is Not a Setting

In January, a federal judge in New York ordered OpenAI to hand over twenty million ChatGPT conversations to lawyers in a copyright case. The logs are de-identified, and the order came with protections. But sit with what actually happened: millions of people told a machine things they may never have told another person, and those conversations became discoverable evidence in a lawsuit none of them are party to, under rules none of them agreed to, decided in a jurisdiction most of them have never set foot in.

No setting could have prevented that. There is no toggle for "my words stay out of other people's litigation." And that's the point I want to make, because it's the quiet assumption underneath almost everything I build: privacy is not a setting. It is a property of architecture. By the time you reach the settings page, nearly every decision that matters about your data has already been made — where it lives, what law it lives under, who can compel it, and what it can be used for by default.

## The setting was never yours

Consider how the last few months have gone for the ordinary user.

In November, LinkedIn began using member data to train its AI models across Europe and beyond — turned on by default, opt-out buried in preferences. The pattern is familiar by now: the choice arrives after the decision, framed as control, functioning as consent theater. You weren't asked whether the system should work this way. You were asked whether you'd like to click a box after it already did.

And even when a company wants to make a real promise, jurisdiction overrules it. Last summer, a Microsoft executive was asked under oath, in the French Senate, whether he could guarantee that French citizens' data would never be handed to American authorities without France's consent. He said he could not. That testimony did more for public understanding of data sovereignty than a decade of policy papers, because it made the structure legible: the data was in France, the servers were in France, the contract said all the right things — and none of it mattered, because the company that operates the machines answers to a different sovereign.

A setting is a promise made inside an architecture. If the architecture reports to someone else, the promise is decoration.

## Governments stopped believing promises first

Watch what the customers with the most leverage on earth are doing about this.

The European Union's Data Act became applicable in September, giving users the right to move between clouds. Amazon launched a legally separate European sovereign cloud in January. In April, the European Commission put roughly 180 million euros into procuring sovereign cloud capacity, and France announced it would begin migrating government workstations off American operating systems entirely — following Denmark, which spent the past year moving its ministries off Microsoft. These are not gestures. They are states concluding, after years of adequacy decisions and privacy frameworks and contractual clauses, that the only guarantee that survives contact with another country's courts is physical and architectural: our data, on our infrastructure, under our law.

I think they're right, and I think the conclusion runs deeper than governments have yet followed it. Because the same logic applies at every scale below the nation — to a hospital, a record label, a research lab, a family. If sovereignty over data can't be granted by promise, only built by design, then it has to be designed in everywhere, not just at the level of national clouds.

## Design it in, or you don't have it

This is the thread that runs through everything I work on, mostly without being announced.

[Tenzro](https://tenzro.com) moves computation to where data already lives, instead of moving everyone's data to where the computation is. A distributed network of nodes means a clinic, a studio, or a ministry can run serious AI on infrastructure it controls, under the law it actually lives under — the model travels, the data doesn't have to. Ipnops approaches the same principle from the other side: as AI reaches beyond the web into the world's data, every access happens under explicit, verifiable rules — who may reach what, for which purpose, on whose terms. And [Vestigim](/essays/vestigim/) carries it into the creative industries, where the people who make the culture AI learns from deserve data they can verify and terms they can see, [a chain of custody](/essays/data-has-a-chain-of-custody-now/) rather than a scraping and an apology.

None of these are privacy products. That's deliberate. Privacy as a product is a setting with better marketing. What I'm after is the structural version: systems where the default is that data stays under the governance of the people it belongs to, and anything else requires permission — instead of systems where extraction is the default and privacy is the paperwork.

## Sovereignty runs all the way down

Sovereignty in the age of AI now means compute, data, and model access — and open infrastructure is the only version of it that doesn't require being rich. That's an argument I usually make at the scale of nations, but it's the same argument at every scale below one. A country renting its intelligence layer from someone else's territory, an artist whose life's work trains a model she can't audit, a person whose private conversations surface in a stranger's lawsuit — these are the same structure at different sizes. Somebody else's architecture, somebody else's jurisdiction, somebody else's default.

The fix is the same at every size too, and it is not a better settings page. It is architecture that makes the promise physically true: computation that comes to the data, access that is governed and verifiable, infrastructure that answers to the people who depend on it. Privacy, done properly, isn't something you configure. It's something you can stop thinking about — because it was decided in your favor before you arrived.
