Token Runner
TOKEN RUNNER // THE A.I. UNIVERSE EXPLAINED · 31 October 2026

Who pays for government when wages shrink

Machines don't pay income tax. Five of every six dollars Washington collects come from people's wages. Ask what happens to schools, pensions and hospitals if the wages go to the machines.

In short

Five of every six federal tax dollars come from income tax and payroll tax, while the tax on company profits brings in less than one in ten.

Workers' share of national income is at or near its lowest since records began, and Yale's model says AI would lift federal revenue only about half as much as it would if workers kept their share.

The safety net already misses most of the jobless, and neither answer on the table, paying people directly or taxing the machines, is law in America.

The question under the layoff

When a company announces a layoff, the story is about the people who lose their jobs. That is the right place to start. But there is a second story under it, and most people never see it, because the money leaves their pay slip before they do.

Every job pays tax. So we went looking for the money. Where does Washington's come from? Which way was it already moving? What do the careful forecasters expect? And what is in place for the people who lose the wage?

Here is what we found. The federal government runs on wages. Workers' share of the nation's income was falling long before generative AI. The best model we found says revenue still grows, but by about half as much. And the safety net already misses most of the jobless.

A government built on wages

In fiscal 2025, the government's accounting year, about half of everything the federal government collected came from individual income tax. Another third came from payroll taxes, the deductions for Social Security and Medicare. Together that is about five dollars in every six. Corporate tax, the tax on company profits, brought in less than one dollar in ten.

OPEN THE INSTRUMENT →

Source: Office of Management and Budget, Historical Tables, Table 2.2, Budget of the U.S. Government, Fiscal Year 2027, https://www.whitehouse.gov/wp-content/uploads/2026/04/hist02z2_fy2027.xlsx (listed at https://www.whitehouse.gov/omb/information-resources/budget/historical-tables/); US government work; read 2026-10-29; raw file raw/omb-hist02z2/2026-10-29T173122.xlsx, sha256 9e2bc741ca3a92929551ca65c6db55901a1d96bac0cae43788c89cfac4f0fe55

OPEN THE INSTRUMENT →

Source: Office of Management and Budget, Historical Tables, Table 2.2, Budget of the U.S. Government, Fiscal Year 2027, https://www.whitehouse.gov/wp-content/uploads/2026/04/hist02z2_fy2027.xlsx (listed at https://www.whitehouse.gov/omb/information-resources/budget/historical-tables/); US government work; read 2026-10-29; raw file raw/omb-hist02z2/2026-10-29T173122.xlsx, sha256 9e2bc741ca3a92929551ca65c6db55901a1d96bac0cae43788c89cfac4f0fe55

Pensions and hospitals run on the pay slip

Social Security is funded almost entirely by the payroll tax: 91.3 percent of its revenue. Its retirement fund, the reserve that tops up what the tax brings in, is already projected to run dry in late 2032. Medicare's hospital fund is projected to run dry in 2033. Those dates were set before anyone modelled AI.

So the question under every AI layoff is not only what happens to the worker. It is what happens to the money for schools, pensions and hospitals if income moves from wages to profits.

The share is already falling

By the government's own measures, workers' share of the national income is at or near the lowest since records began.

The Bureau of Labor Statistics measures the share of business income paid to workers. In the second quarter of 2026 it stood at its lowest since 1947. The Bureau of Economic Analysis measures employee pay as a share of national income. In 2025 that was 50.9 percent, the lowest since 1935. In 1970 it was 58.4 percent.

That shows the fall. It does not prove AI caused it. The decline began long before generative AI, with globalisation, weaker unions and the rise of software. But it tells you which way the money was already moving when the machines arrived.

OPEN THE INSTRUMENT →

Source: U.S. Bureau of Labor Statistics (Productivity and Costs) and U.S. Bureau of Economic Analysis (Gross Domestic Product), via FRED, Federal Reserve Bank of St. Louis: https://fred.stlouisfed.org/series/PRS85006173; https://fred.stlouisfed.org/series/A4002E1A156NBEA; US government works; read 2026-10-29; raw raw/fred-labour-share/2026-10-29T154354/PRS85006173.csv sha256 f3b9ce09c3d14893e94e7eeaffd4074f6598df7dcd893ba52335198bf65fd3ce; raw/fred-labour-share/2026-10-29T154354/A4002E1A156NBEA.csv sha256 9069523b6fdbdcc83c477df09ba90850013d38d5fa3a540f30ac2b7abd58b9dc

OPEN THE INSTRUMENT →

Source: U.S. Bureau of Labor Statistics (Productivity and Costs) and U.S. Bureau of Economic Analysis (Gross Domestic Product), via FRED, Federal Reserve Bank of St. Louis: https://fred.stlouisfed.org/series/PRS85006173; https://fred.stlouisfed.org/series/A4002E1A156NBEA; US government works; read 2026-10-29; raw raw/fred-labour-share/2026-10-29T154354/PRS85006173.csv sha256 f3b9ce09c3d14893e94e7eeaffd4074f6598df7dcd893ba52335198bf65fd3ce; raw/fred-labour-share/2026-10-29T154354/A4002E1A156NBEA.csv sha256 9069523b6fdbdcc83c477df09ba90850013d38d5fa3a540f30ac2b7abd58b9dc

What the modellers say

The careful forecasters do not say the tax base collapses. Yale's Budget Lab modelled AI futures under the current tax system. In the fastest case, federal revenue still rises, by up to $246 billion in fiscal 2030.

But that is only about half of what the same growth would bring if workers kept their share. The reason is that capital, the money made from owning things, is taxed more lightly than work. As AI widens the gap between owners and workers, payroll tax revenue falls.

The money does not vanish. It moves from wages, which are taxed heavily and at source, to profits and gains, which are taxed lightly and often later.

What a town loses

The national numbers hide the local ones, and the local loss is the easiest to see. When a large employer leaves, a town loses more than wages. It loses income tax, sales tax, property values, and the shops that depended on the workers.

We have not found a town hollowed out by a layoff its employer blamed on AI. But we know what one looks like. When Tyson closed its beef plant in Lexington, Nebraska, in January 2026, the University of Nebraska put the loss at 7,003 jobs. That meant about $3.28 billion a year in output, and $23.2 million a year in local tax alone.

"Tyson owes this community a debt," the city manager said, as the AP reported.

The net already has holes

Before any forecast, there is a measured fact. America's safety net for the jobless already misses most of them.

Unemployment insurance is the weekly payment a laid-off worker can claim. The St. Louis Fed found in March 2026 that the share of the unemployed eligible for it ranges from 30.5 percent in Alaska to 70.3 percent in New Hampshire. Benefits replace between 42.6 and 67.1 percent of lost wages, depending on the state. The share of unemployed people who actually receive benefits fell from more than half in the 1950s to less than 30 percent, according to the Washington Center for Equitable Growth. Before a single AI layoff, most of the jobless are already outside the system.

In America, a lost job often means lost health insurance too. The Commonwealth Fund projects that the expiry of the enhanced health-insurance tax credits will cost about 339,100 jobs in 2026, and the Urban Institute estimates 4.8 million people will become uninsured. Losing the subsidy costs jobs, and losing the job costs the coverage.

People are already reaching into their retirement savings. Hardship withdrawals from 401(k) plans, the workplace retirement accounts, hit a record in 2025: 6 percent of participants, up from about 2 percent before the pandemic, in Vanguard's figures as Axios reported them.

Senator Ron Wyden announced a bill on 23 October 2026 to widen unemployment insurance for workers hit by AI. We print no number for it.

Answer one: pay people directly

Universal basic income, a regular payment to everyone with no conditions, came back into the conversation in 2026, pushed by officials more than by chief executives. The evidence supports neither side as cleanly as each claims.

In the biggest US study, 1,000 people received $1,000 a month for three years, against 2,000 who received $50. They worked about 1.3 hours a week less and earned somewhat less from work. The lab that ran it says those differences are not statistically significant. Recipients rated work as more important, not less. Its funder was the chief executive of OpenAI. In Kenya, total work did not fall at all: people moved from wage jobs into their own businesses.

Every pilot ends, and almost none follows people afterwards. Wales is the exception. Its government published what happened when the money stopped: savings went quickly, debts and arrears returned, and some recipients found the ordinary benefit system more constraining than before.

Source: OpenResearch, Unconditional Cash Study (NBER w32719); Banerjee and others, the GiveDirectly Kenya study; AEI, Corinth and Mayhew (25 February 2026); each row names its source

The problem is the bill. Paying every American adult $12,000 a year would cost about $3 trillion a year, two economists, Hoynes and Rothstein, estimated in 2019. That is about three quarters of all federal spending, before any savings. We found no official costing of a national basic income.

Answer two: tax the machines

In 2017 Bill Gates suggested that a robot doing a worker's job should pay tax like the worker. In August 2026 he went further. "The tax system nudges you toward replacing people with machines," he said, as Fortune reported it.

That same month the idea reached bill text. H.R. 10044, the AI Tax and Work Protection Act, would tax AI usage at a rate that rises automatically with unemployment. It would include a company's own use of AI where it "enables or results in a reduction in the workforce."

South Korea tried a version. In 2018 it cut the tax break on automation equipment, and a 2025 study found affected firms bought fewer robots and hired more people. The European Parliament voted the idea down in 2017.

Neither answer is law in America, and each has its critics. Geoffrey Hinton, the Nobel laureate who expects mass unemployment, still rejects cash as the answer: it "won't deal with human dignity," he told the Financial Times. Senator Mike Rounds, a Republican, put the objection to the tax plainly, as Roll Call reported in June 2026: "If you start to tax it, you slow it down, and it goes elsewhere."

The builders propose the safety net

There is an odd feature of this debate. Some of the loudest voices for new taxes and new payments are the people building the machines. OpenAI published a paper on taxing capital. Elon Musk spoke of "universal high income," as reported. Anthropic's chief executive wrote an essay on tax, and has been reported as suggesting a tax on AI model revenue.

Disclosure: Token Runner is built with Claude, which is made by Anthropic.

Whether that is conscience, insurance or positioning, it means the question of who pays is no longer fringe. By October 2026 Congress had rival bills. The argument had moved from "should we" to "how."

Why it matters

There are two ways to read this, and the evidence supports both.

The first is that the tax base is quietly eroding. Washington runs on wages. Workers' share of income is already near a record low. The pension and hospital funds are already short. If AI moves more income from wages to profits, a government built on wages has to change how it collects, or spend less on the people who lost them.

The second is that the alarm runs ahead of the numbers. The careful model does not show revenue falling; it shows revenue rising, by less. The fall in workers' share began decades before generative AI. No town has yet been hollowed out by a layoff blamed on AI. And some of the loudest calls for new taxes come from the companies building the machines, whose reasons are their own.

Either way, the jobs story gets the headlines, and the tax story decides what happens next.

Our ask is simple. When the next layoff is announced, ask two questions: who lost the job, and who now pays the tax that job used to pay?

Sources