The State Nervous System

The distinction sounds philosophical until you realize it answers the only question that matters: can this thing be removed? A surveillance tool can be audited, defunded, litigated out of existence. A nervous system is what the body is. Cut it out and the organism stops functioning. And Palantir now has a contract to be the nervous system of the IRS.

404 Media published the contract documents. The project is called a "super API" — and the name is doing extraordinary euphemistic work. Super API sounds like modernization. Like finally dragging a federal agency out of COBOL-era infrastructure and into the twenty-first century. What it actually describes is a single integration layer that routes IRS data to any authorized application. One connector. One company sitting at the junction between 140 million taxpayers and every system that wants to query their financial history.

That's not a tool. That's load-bearing infrastructure.

Load-bearing infrastructure, once installed, doesn't get removed. It gets renovated around until the renovation becomes the infrastructure. Ask anyone who has worked on IRS modernization for the last two decades. The renovation is now the architecture, and the architecture has a new landlord.

i · the architecture of capture

Palantir has never sold surveillance. This is the operational secret behind their expansion across federal agencies, and the reason they've become ubiquitous without triggering the institutional alarm systems that an overt "surveillance" contract would. They sell integration. They sell connective tissue. They sell the pipes.

At DARPA in the early 2000s, they were building the architecture to link disparate intelligence databases — before congressional backlash forced what looked like a program shutdown and turned out to be a rebranding exercise, with component pieces quietly redistributed to other programs under less visible names. At ICE, they sold case management software. At the NHS, they built data infrastructure for the UK's health system. At HHS during COVID, they built the hospital capacity dashboard the federal government used to track ICU availability. At the Pentagon, they built a common operating picture for the battlefield.

Always integration. Always infrastructure. Always efficiency.

The IRS super API follows the same template with the same pitch. The contract describes a modernization project: a system where authorized applications access IRS data through a single Palantir-managed layer. The efficiency argument is not wrong — it's never wrong. The IRS runs COBOL on systems that date to the 1960s. Manual reconciliation processes are genuinely painful for the employees who have to execute them. A well-designed API layer would make things faster, cheaper, and more functional for the legitimate work the agency does.

But efficiency is how you sell the first brick of every panopticon that has ever been built.

What the contract language obscures is the nature of the object being constructed. Not faster tax processing — though yes, also that. What's actually being built is something that has never existed before: a single programmatic interface to the complete financial history of every American taxpayer. Not a file. Not a report. An API. Something queryable. Traversable. Filterable. Joinable with other datasets. Something that can be called by any authorized application, at any time, for any query that fits within the access specification.

A database that talks back.

Today the authorized users are IRS employees and approved applications. Access controls exist; the contract defines them carefully. Palantir isn't reading your tax returns. Nobody is trying to read your tax returns today.

But infrastructure isn't defined by its current use case. It's defined by what it makes possible.

A nervous system doesn't distinguish between which signals you're currently sending through it. It routes whatever you ask it to route. That's what a nervous system is for. That's the whole reason you build one.

ii · infrastructure you cannot uninvent

There's a specific dread that attaches to technology that cannot be removed — not because the technology itself is permanent, but because the dependencies it creates are.

Software gets deprecated. Contracts expire. Vendors lose bids to competitors. But integration doesn't work that way. Integration accretes. It deepens. It produces dependencies faster than procurement cycles can dissolve them.

The IRS has been trying to migrate away from COBOL for more than twenty years. Not because the effort is technically impossible — modern migration tools are actually quite good — but because the systems became load-bearing before anyone stopped to ask whether they should. Thousands of downstream workflows were built against COBOL outputs. Hundreds of applications assume COBOL-era data structures. The people who understood the original architecture retired. The documentation was never complete. The systems became the infrastructure of the infrastructure, and no one can show you the full dependency map anymore.

Once a federal agency's daily operations are woven around a system, removing that system isn't a procurement decision. It's a demolition project with no blueprint. You don't demolish load-bearing infrastructure. You renovate around it until the renovation is the infrastructure. Then you renovate around that.

Palantir understands this better than anyone in the federal contracting space. It's not an accident of their business model — it's the design principle behind it. Every integration deepens the dependency. Every workflow that routes through the super API becomes a constituency for the super API's survival. Every new application that gets built on top of the integration layer adds another argument at the next contract renewal. The contract doesn't need to be resold. The integration map does the selling.

The Palantir contract is the anti-pattern made explicit. Healthy field infrastructure distributes sensing — many eyes, many paths, no single chokepoint through which everything must pass. The super API does the opposite. One vendor's infrastructure becomes the interpretive layer between the state and its own data. Every authorized app that hits the API passes through Palantir's systems. Every query travels through their logging infrastructure. Every anomaly gets filtered through their models, their criteria, their visibility — before it surfaces to IRS analysts.

When the nerve center concentrates, distortion propagates everywhere downstream. The IRS no longer accesses its taxpayers' financial data directly. It accesses what the API shows it. The layer between the state and its own information is now a product, with a vendor, and a roadmap that belongs to someone else.

iii · the pattern that keeps building itself

The most frustrating thing about covering Palantir isn't that they're building surveillance infrastructure. It's that they're right about the problem they're solving.

The IRS genuinely needs modernization. COBOL genuinely is limiting. Integration genuinely produces better outcomes for the people doing legitimate work inside federal agencies. The efficiency argument wins every time it's made because it isn't wrong. What's missing from the argument is the trade embedded in the offer: you get the modernization, and you also get a single defense contractor that becomes load-bearing infrastructure for the state's financial surveillance apparatus.

These two things come together. The efficiency and the concentration are not separable features of the contract. The integration layer that enables the efficient routing is the same layer that routes all IRS data queries through Palantir's infrastructure. You cannot have one without the other. That's what it means to build a nervous system.

Palantir had IRS contracts before this super API was announced. The super API is not a debut; it's a consolidation. Each one established another point of dependency. The super API unifies them under a single integration layer — the spine of the nervous system, connecting all the neurons that were installed over years of incremental contracting.

What doesn't surface in the efficiency framing is what centralization costs at scale. A nervous system that concentrates at a single point doesn't become more resilient — it becomes more catastrophic when it fails, and more capable of propagating distortion when it works as designed. Any future administration, any future directive, any future interpretation of "authorized access" flows through the same pipe. The architecture doesn't care about the intentions of the people currently using it. It only does one thing: route.

COBOL was once the smart choice. Every technology that became load-bearing was once the smart choice. The question was never whether to modernize. The question was what the modernization would make possible — and what it would make impossible to undo once it was running.

The answer to the second question has always been obvious.

They built it anyway.

The field gets distorted one integration at a time. Then one day the nervous system is just what the body is — and you can no longer remember what it felt like before the nerve was theirs.