EP 04 Season II · May 2026 opening the arc to the season finale

The Influence War

They don't have to change your mind. They just have to catch you in the right feeling.

Read time

17 minutes · 9 sections · one interactive

Position

Fourth of six episodes in Season II · bridges from commerce to civic

Companion

Sets up EP 05 · The Composite State and the season-closing Convergence essay

Begin reading

S 02 · The acceleration

More ways to see you. More often.

Two things have happened in parallel for a century. The number of ways government can observe a citizen has grown — slowly at first, then in layers, then in compounding bursts. And the frequency at which each observation happens has shifted from periodic to continuous.

For most of its modern history, the surveillance question was what is technically possible. Each new capability arrived as its own moment — fingerprinting in 1910, federal records in 1935, computerized warrants in 1975, mobile location in 2005, real-time biometric face matching after 2010. Each became a layer on top of the previous, and each layer thickened.

What changed in the last decade is not only new capability — it is frequency. Every existing capability now runs continuously, and the data flows into cross-referenced storage rather than isolated paper files. The same persuasion logic that learned to read commercial mood now works in civic settings too: the system that reads the receiver's mood is substantially more successful than the system that does not.

The surveillance stack · what became possible when

civic data collection capabilities · year of first scaled deployment · darker right = more continuous now

each row is a layer the state could add on top of the previous · the stack has grown taller and denser without subtraction

Every row represents a capability that, once deployed at scale, has never been removed. They accumulate. Each new row sits on top of the prior twelve and the architecture compounds. The stack only grows.

What follows is the rate at which both collectors — government and commercial — have grown alongside each other. The shape of the two trajectories is the entire editorial story.

Two collectors. Same century. Different trajectories.

data collection capacity · 1900 → now · stylized · government grew first, commercial overtook, AI is closing the gap

civic · government data collection (census, IRS, SSA, federal records, surveillance) commercial · ad-tech, brokers, platforms, wearables

The line is the funded surface of every fragment of advertising you have ever seen. It runs flat for decades. Then television arrives. Then cable. Then the open web. Then social. Then microtargeting. Then the wearables.

The shape is not an accident. Each capability layer compounds the last. The system is not louder in 2026 than it was in 1956. It is more precise.

Until the system could see your mood, persuasion was a guess. Once it could see your mood, it became arithmetic. And arithmetic compounds.

So — is the system still an event you could point to?
Or has it quietly become infrastructure: always on, rarely noticed?

section 02 of 06 · the acceleration · what became possible when Section 03 · The convergence

S 03 · The convergence

Two machines. One room.

Everything you've read so far has been one machine — the commercial one. There is a second machine, growing alongside it for decades, that collects you under entirely different rules. Until very recently the two lived in different rooms. The hyperscale era put them in the same room — and put an AI inside that can read both.

While the commercial machine has been industrializing for a century, another machine has been growing alongside it — quieter, less visible, no less complete. The state collects you too.

It collects a different set of data, for a different purpose, under different rules. But it collects continuously and — under the right legal authority — connectably.

Before the analytics, watch it happen. Scroll, and you assemble.

composite assembly · scroll to build · ochre = commercial · plum = civic · amber = the bridge

01 · the frame

This is your profile. Unbuilt.

An empty frame, split down the middle. Two machines are about to fill it — each building a different half of you. Scroll, and watch them work.

02 · the commercial half

The commercial machine builds one half.

purchases · location · browsing · social · mood · wearable — one half of you, assembled to be sold to. The broker composite from Episode 02 pours into the left half like ink into water.

03 · the civic half

The civic machine builds the other half.

IRS · DMV · court · benefits · voter roll · surveillance — the other half, assembled to be governed. Different data, different purpose, pouring into the right half.

04 · the scan

Now the AI scans the whole.

This is new — only a few years old, the coverage still emerging, not yet fully realized. A third color floods the outer ring, the dividing lines fall away, and a single scan begins reading across both halves at once — what no human ever assembled.

05 · convergence

The colors blend into one.

Ochre, plum and amber resolve into a single new color — a state that did not exist at the start. The empty frame is now a filled circle. Two machines, one scan, one answer. This is you, whole — to a system that until recently could only ever see you in pieces.

— Commercial

prediction.

"what will move you?"

purchase history broker
browsing patterns cookies/SDK
location pings app
social graph platform
inferred income modeled
engagement signals platform
wearable telemetry opt-in
inferred mood modeled

→ governed by terms-of-service · weakly enforced · commercial discretion

— Civic

authority.

"what are you, legally?"

tax filings IRS
court records PACER · state
DMV / license state
property deeds county
benefits enrollment agency
voter registration state · SoS
traffic / ALPR municipal · state
surveillance footage MPD · transit
body camera archives law enforcement
intelligence records federal

→ governed by statute · subpoena · FOIA · no opt-out

If knowledge is power, this is a power dynamic worth naming. Two collectors. Both have been accumulating knowledge of you for a century. The public conversation has focused almost entirely on one of them.

The other has been growing quietly the entire time — and is now in a position to use the same hyperscale infrastructure to catch up.

The absence of comprehensive U.S. privacy regulation has often been read as oversight. It may also be read as strategic delay: a regulator does not constrain capabilities they themselves expect to use.

The commercial machine wants to predict you so it can sell to you, target you, persuade you. The civic machine wants to fix you — in the sense of fix in place: identify you, tax you, license you, enforce against you, deliver benefits to you, hold you accountable to a record. Different verbs. Different rooms.

For most of their history, these two pools lived in different infrastructure. Different vendors. Different storage. Different access patterns. The integrator that could read across both was either technically impossible or so legally and operationally expensive that no one attempted it at scale. The friction of integration was its own informal protection.

That changed in the hyperscale era.

AWS GovCloud sits in the same data centers as AWS Commercial · Azure Government runs the same architecture as Azure Public · same buildings · same handful of corporate roofs · one AI reading across both partitions

Three companies host most U.S. cloud workloads. The same three companies host most U.S. government cloud workloads. AWS GovCloud sits in the same data centers as AWS Commercial. Azure Government runs the same architecture as Azure Public. Same buildings. Same power. Same fiber. Same handful of corporate roofs.

On top of that physical convergence sits a layer that wasn't possible ten years ago: AI that reads across silos. The same model can be pointed at the commercial composite and the civic composite. Same vector embeddings. Same retrieval architecture. The integrator that used to require enormous legal and operational expense is becoming a contract clause and an API call.

The friction was the protection.

1995 background check · standard query

01DMV records · paper return · mail3–7 days

02Subpoena court records · clerk processing5–14 days

03Credit pull · written authorization1–2 days

04Cross-agency phone calls · variable2–5 days

05Human review of compiled file1–3 days

total 14–30 days

2024 same query · AI-augmented

01Single API call to data aggregatorinstant

02Cross-source AI join · automaticinstant

03Structured return with confidence scoresinstant

total ~ 4 seconds

The slowness wasn't inefficiency. It was review windows, mistake-surfacing time, appeal opportunity, negotiation space. The friction was the protection — and the protection has been removed without being replaced.

— What happens to "justice" when justice becomes swift?

Historically, the administrative state operated at human pace. A clerk pulled a file. A second clerk verified it. A supervisor signed off. The process took days or weeks — not because anyone wanted it slow, but because the protections were embedded in the steps themselves. The pace was the due process.

AI-augmented government workflows compress that timeline to seconds. ALPR plus warrant database equals immediate pursuit. Predictive risk scoring equals pre-arrest classification. Automated benefits determinations equal instant approval, instant denial, instant appeal denial. Cross-agency profile assembly means no human ever assembled the file — but the file decides what happens to you.

When justice becomes swift, the felt experience of being governed changes. The question is not whether the AI is more accurate — that is contested. The question is whether the protections built around slow processes survive their speed-up. So far, the protections have not been rebuilt at the new speed.

This is not a forecast about something that might one day arrive. It is a map of decisions already being made — country by country — some pushing the capability forward, others drawing a legal line against it. The point is not that the worst outcome is inevitable. The point is that the outcome is open, and it is being decided now.

field evidence automated & predictive policing — where it already operates, and where it's already restricted

US United States Predictive-policing systems (PredPol / Geolitica) adopted by roughly 40 law-enforcement agencies since 2012; automated license-plate-reader networks operate nationwide.AP / MIT Technology Review reporting, 2012–2023 deployed

CN China Facial-recognition patrols and integrated police data platforms operate at national scale, fusing identity, movement and behavioral records.Human Rights Watch / academic field studies deployed

UK United Kingdom Police forces have piloted algorithmic risk-scoring tools (e.g. Durham's HART) to classify individuals by reoffending risk before charge.The Law Society / academic audits piloted

IN India Cities including Hyderabad have run facial-recognition and predictive-surveillance pilots integrating CCTV, ID and police databases.Internet Freedom Foundation reporting piloted

NL Netherlands Predictive-policing and welfare-fraud risk systems were deployed, then partly halted after courts and audits found rights violations.District Court of The Hague, SyRI ruling 2020 restricted

EU European Union The AI Act (Article 5, in force February 2025) prohibits AI that predicts criminal behavior on the basis of profiling alone.Regulation (EU) 2024/1689 restricted

Two facts hold at the same time: the capability is real and already running in some places — and elsewhere it has already been written into law as a line not to cross. The future here is not fixed. It is being chosen, jurisdiction by jurisdiction.

The persuasion machine you just walked through and the administrative state you live inside are no longer in different rooms. They are bridgeable in the same room — by the same intelligence — for the first time in human history.

That bridging hasn't happened comprehensively yet. There is still legal friction, internal-policy friction, contractual friction. But the technical friction — the friction that used to do most of the protective work — is gone. What remains is whether anyone notices, and whether anyone chooses to stop it.

echo · this is no longer hypothetical

In 2025, a version of the bridge was assembled in public — and at speed.

≈250,000

NET FEDERAL WORKFORCE REDUCTION · 2025

mostly voluntary exits · not a layoff count

UNITED STATES · FEDERAL CIVILIAN WORKFORCE

↻ ZOOMING OUT FROM DEKALB → THE NATIONAL PATTERN

The metro-Atlanta displacement story doesn't stay local. Across 2025 the federal civilian workforce shrank by roughly 250,000 on net — the largest peacetime reduction on record — draining from hubs nationwide, concentrated in the Washington, DC metro. Read this carefully. This is a net headcount change, not a layoff count. Public OPM and GAO figures show the overwhelming majority left through voluntary channels — the Deferred Resignation Program (the “fork in the road,” ~75,000+ early takers), Voluntary Early Retirement (VERA) and buyouts (VSIP). Involuntary reductions-in-force were a small share — under ~1% of separations at the major agencies in the first half of 2025. The figure does not isolate forced cuts from early retirements, voluntary resignations, transfers, or moves to new opportunities.
↳ The mechanics of AutoRIF & GSAi · Episode 05

Early in 2025, the Department of Government Efficiency (DOGE) moved to consolidate access across federal systems and apply AI tools to decisions that used to run through layered human review. The same pattern this episode describes — collected data, joined by a model, acting at machine speed — appeared inside the government itself.

~250knet drop in the federal civilian workforce across 2025 — mostly voluntary exits — thinning the human review layer

AutoRIFa legacy personnel tool reportedly repurposed with AI to help drive mass-termination decisions

GSAian internal AI assistant rolled out to roughly 1,500 federal workers; agency data flowing toward LLM tooling

Read it next to the metro-Georgia thread that runs through this series — counties whose residents move, get re-sorted, and re-registered inside civic databases (the displacement story behind DeKalb and its neighbors). The local machine collects. The federal initiative supplies the model and the speed. Put them in the same room and you don't need a new law to change what's possible — you only need an account and an API key.

None of this proves intent. It proves capacity: the components are now assembled and operating. Whether they're pointed at efficiency, at oversight, or at something else is the open question the rest of this season takes up.

Sources: OPM workforce-changes data and GAO reporting on 2025 federal workforce reductions (composition: ~92%+ voluntary programs, <1% RIF, Jan–Jun 2025) · AP, Reuters, Wired and ProPublica on DOGE operations and GSAi deployment, 2025 · Infera series, metro-Atlanta displacement thread

Where this episode ends, the season pivots. The commercial trilogy (Episodes 02, 03, 04) describes the machine that wants to move you. The civic episodes (05, 06) describe the machine that wants to fix you in place. And the convergence is the third subject — what these two machines, now in the same room, could become together.

before 2010

two machines · two infrastructures · no integration

2010 – 2018

cloud consolidation · same vendors emerge on both sides

2018 – 2023

physical co-location · GovCloud + Commercial · same buildings

2024 → now

AI reads across both · integration becomes an API call

section 03 of 06 · the convergence · two machines, one room Section 04 · The counter-read

S 04 · The counter-read

Three voices, on the record.

A technologist who welcomes it. A senator who fears it. A scholar who named it. Each quote is published and sourced — and none of them speaks for the publication.

Larry Ellison

Co-founder & CTO · Oracle

The technologist case is that pervasive recording produces better behavior. In a 2024 financial analyst meeting, Ellison described an AI-supervised society in approving terms — constant monitoring as a feature, not a cost.

"Citizens will be on their best behavior, because we are constantly recording and reporting everything that is going on."

— Oracle Financial Analyst Meeting, Sept 2024

Bernie Sanders

U.S. Senator · Vermont

The political warning is that the pace of capability has outrun the pace of oversight. Sanders has introduced legislation to pause the AI-data-center build-out until Congress can establish safeguards.

"We cannot sit back and allow a handful of billionaire Big Tech oligarchs to make decisions that will reshape our economy, our democracy and the future of humanity."

— Sanders & Ocasio-Cortez, AI Data Center Moratorium Act (S.4214), 2026

Shoshana Zuboff

Professor Emerita · Harvard Business School

The scholar frame is that the asymmetry is the point: the knowledge accrues to the systems, the loss accrues to us. She coined surveillance capitalism to name a mechanism most people experience before they can describe it.

"The knowledge that now displaces our freedom is proprietary. The knowledge is theirs, but the lost freedom belongs solely to us."

— The Age of Surveillance Capitalism, 2019

Where the publication stands

These three do not agree with one another. Ellison welcomes the supervised society; Sanders wants to slow it; Zuboff wants to name and limit it. Hosting all three is the point — the convergence is real enough that a technologist, a senator, and a scholar are each already on the record about it.

The publication position is narrower than any of theirs: not that the worst outcome is inevitable, but that the technical friction that used to slow this down is gone, and the question of what replaces it is still open.

section 04 of 06 · the counter-read · hosted fairly Section 05 · Discipline

S 05 · Discipline

What we did not prove.

Four restraints. Held visibly. The quietest section of the episode is the one that earns the rest.

✓

Restraint 01

We did not prove that any specific election was decided by microtargeting. The infrastructure exists; the question of how much it changes any specific outcome is contested and should remain so.

✓

Restraint 02

We did not prove that any individual voter changed their mind because of a single ad. The ecological-fallacy guard is non-negotiable. The reader's mind is theirs.

✓

Restraint 03

We did not prove that personalized persuasion is more powerful than mass persuasion. We documented that it is more continuous, more granular, and more auditable in its inputs — and that those properties compound differently than mass-broadcast persuasion ever could.

✓

Restraint 04 · Load-bearing

The system does not mind-control anyone. It optimizes the emotional conditions under which certain beliefs, reactions, and identities are more likely to emerge. The reader's agency remains the load-bearing variable. Anything tighter than that drift is intellectually sloppy and gives defenders an easy out.

what we did prove

The infrastructure for industrial-scale identity targeting exists, is in continuous active use by both U.S. parties and most major issue campaigns, and its inputs are the broker composites and trust scores documented in Episodes 02 and 03.

section 05 of 06 · discipline · four restraints Section 06 · Closer

S 06 · Closer

The most effective persuasion systems
do not change your opinion. They change what feels true.

And the harder question the convergence leaves open: when both machines finally share one room — can the government control what it has built? Does it want to? Or is it, like the rest of us, still only trying to understand it?

— The Influence War · Episode 04 · The Mechanism Series

✦

A reward · for the readers who reached the end

Most people don't make it this far.

So here's something built only for the ones who did — a single-sheet Case File that folds this whole episode into one artifact you can keep, print, or hand to someone who'll never read four thousand words.

click to unseal

The Influence War · Case File.pdf

one page · the whole machine, on a single sheet

⬇ Download the Case File

Yours to keep. No email, no signup, nothing tracked. The receipt the other thousand systems never give you.

What comes next

The commercial machine takes your attention. The state version takes your authority. EP 05 · The Composite State looks at the parallel machine the government runs — and where it physically converges with the commercial one.

Continue to Episode 05