Three things happened in April and May of 2026 that, taken separately, look like unrelated news stories. Taken together, they describe a single pattern: AI infrastructure is being built at every level of government and commerce, and democracy is not in the room where it happens.
The Federal Layer
The Drey Dossier, an independent investigative channel, published a thesis in late April that the White House ballroom construction project (the one with the $17.4 million no-bid contract for Lafayette Park fountains, five times the independent estimate) is cover for a classified AI data processing facility. The donor list reads like a defense procurement manifest: Palantir, Booz Allen Hamilton, Lockheed Martin, Amazon, Google, Microsoft, Meta. Not hospitality companies. Data infrastructure companies.
The thesis: a hardened, airgapped facility running AI on consolidated federal agency data (IRS, Social Security Administration, DHS, VA health records through Oracle’s Cerner system) beyond congressional or judicial oversight. Funded not by appropriation but by corporate “philanthropy” from the companies that would operate it.
The constitutional question reached the appeals court on June 5. Two of the three judges appeared skeptical of the administration’s claim that no court could order the project stopped; one described the government’s position as “move fast and break things and then nobody has standing.” The panel has not yet ruled. Construction continues.
Congress did not authorize this. No committee voted on it. No citizen was consulted. The facility, if it exists, processes the data of every American who has ever filed a tax return, applied for Social Security, or received care at a VA hospital. The AI does not serve these citizens. It processes them.
The Regulatory Layer
In early May, the White House blocked Anthropic from expanding access to Claude Mythos, its most capable AI model, from fifty organizations to one hundred and twenty. Two reasons were cited: national security, and concern that wider access would strain the government’s own compute allocation. Meanwhile, OpenAI was shipping GPT-5.5 to nine hundred million weekly users. The capabilities one company restricts from release were generally available through the other’s product.
That was the soft version. On June 12, the Commerce Department issued an export control directive ordering Anthropic to shut down access to both Mythos 5 and its commercially released version, Fable 5, for all customers worldwide, citing a claimed jailbreak. Anthropic complied within hours, disabling both models globally rather than attempting to segment access by nationality. It was the first time a U.S. government directive forced a commercially deployed frontier AI model offline.
What began as informal pressure has become a kill switch. There is no statute. There is no appeals process. There is no public record of the technical criteria that triggered the shutdown. The sixty-eight organizations originally denied expanded access, and now the hundred and fifty Glasswing partners who had it, learned about these decisions after they were made. States, cities, school districts, and civic organizations were never part of the conversation at all. The question of who gets to use the most powerful thinking tools ever built is being answered daily, by executive discretion, in the absence of any democratic framework.
The Municipal Layer
And then there is Woven City. Toyota spent an estimated ten billion dollars building a sensor-laden urban environment on a disused factory site at the base of Mount Fuji. One hundred handpicked residents moved in six months ago. Every street is lined with cameras at a density the visiting journalist said he had never seen anywhere in the world.
All those cameras feed into the Woven City AI Vision Engine, an agentic system that monitors, catalogs, and reports activity. It tracks people by their clothing, following individuals across the camera network. Toyota plans to refine this system at Woven City and then sell it to municipalities worldwide.
The residents are not citizens of Woven City. They are test subjects in a product development process. The AI does not serve them. It learns from them — learns how to watch, how to track, how to coordinate vehicles and pedestrians and infrastructure into a single system — so that Toyota can sell the capability to the cities where the rest of us actually live.
When your city gets the pitch — and it will — the sales deck will say “tested in a real city, proven to work.” What it will not say is that the city was a corporate laboratory and the residents were handpicked volunteers with no democratic voice in what was built around them.
The Pattern
Federal. Regulatory. Municipal. Three layers of governance, three ways AI infrastructure is being built without democratic input. The pattern is consistent:
The infrastructure is built first. The governance question is deferred. The people whose data is processed, whose access is restricted, whose streets are surveilled, participate in none of the decisions and learn about them after the fact.
This is not a conspiracy. It is the default. When powerful technology meets weak institutions, the technology gets built by whoever can afford it, governed by whoever has leverage, and experienced by everyone else as a fait accompli. The conspiracies are unnecessary. The incentives are sufficient.
The Alternative
We have spent three years building a civic framework called Foundation on the premise that the AI transition requires new governance infrastructure, not just new technology. Foundation’s answer is not “stop building.” It is “build differently.”
At the energy layer: distributed, community-owned power generation from repurposed infrastructure, not concentrated corporate data centers drawing grid power. Oklahoma’s Data Center Consumer Ratepayer Protection Act (HB 2992), signed into law in May and operative July 1, 2026, already requires large loads to pay their own way rather than shifting costs to residential customers. It is a start, not a solution, but it exists because a state legislature took the question seriously before Washington did.
At the connectivity layer: mesh networks that work without cell towers, without ISPs, without anyone’s permission. Communication infrastructure that citizens own and operate, not camera networks that corporations own and municipalities license.
At the governance layer: a Guardian AI, a publicly owned artificial intelligence operating under democratic oversight, with a published constitutional document, serving deliberation rather than surveillance. An AI whose source code is public, whose instructions are versioned and citizen-reviewable, whose purpose is to help people think rather than to process them as data.
The Guardian does not exist yet. The legislation authorizing it does not exist. The institutions that would oversee it do not exist. But the need grows with every facility built in secret, every access decision made by executive fiat, every camera installed in a corporate test city.
The thinking machines are already here. The question of who they serve is being answered every day, by default, not by choice. Every day without governance infrastructure is a day the answer gets harder to change.
We think it is worth changing. And we think the people who are being processed, restricted, and surveilled without their consent deserve a seat at the table where these decisions are made.
Not after the infrastructure is built. Now.