Who Funds the Safety Net When the Labor Income That Funds It Begins to Shrink?

A worker loses a job.

Unemployment insurance helps replace part of the lost paycheck. Food assistance helps the household buy groceries. Medicaid may provide healthcare. Retraining programs may help the worker prepare for another occupation.

That is what a safety net is supposed to do.

It catches people when their connection to the labor market temporarily fails.

But there is an assumption buried inside that system.

That assumption makes sense when unemployment is cyclical, temporary or concentrated in a limited number of industries.

It becomes much harder to maintain if artificial intelligence begins reducing labor demand across many occupations at the same time.

AI displacement would not only create more people who may need help. It could also weaken the tax base expected to provide that help.

That is the fiscal side of the Paradox of Automation.

Labor Income Does More Than Pay the Worker

When a person receives a paycheck, the economic effect does not stop with the person.

The paycheck produces payroll taxes. It produces individual income taxes. It finances purchases that generate sales and excise taxes. It supports mortgage payments, retirement contributions, insurance premiums and the consumption that keeps other people employed.

Labor income does two jobs at once.

It supports the household, and it supports the public systems surrounding the household.

The United States does not finance its entire government through labor income. Individual income taxes also reach capital gains, interest and business income. Governments collect corporate taxes, property taxes, sales taxes, tariffs and other revenue.

But the system leans heavily on wages and employment.

Over the past fifty years, individual income taxes have supplied nearly half of federal revenue, while payroll taxes have supplied another 35%. Payroll taxes primarily finance Social Security and Medicare Part A.¹

Not every dollar of individual income tax is a tax on wages. But wages are a major part of the base, and payroll taxes are attached directly to employment earnings.

Social Security is social insurance rather than a means-tested safety-net program, but it makes the same tax-base dependence especially visible. In 2025, payroll tax contributions supplied 91.3% of the income received by the combined Social Security trust funds.²

The Safety Net Works in Both Directions

When employment and income fall, government finances automatically move in two directions.

Tax collections decline because households earn less.

At the same time, more people become eligible for unemployment insurance, food assistance and other support.

The Government Accountability Office describes these mechanisms as automatic stabilizers. They help keep a downturn from feeding on itself: tax liabilities fall, benefits rise and households retain at least some ability to spend.³

That is not a defect. It is one of the ways a modern economy absorbs a shock.

But automatic stabilizers were largely designed around a familiar pattern:

What if the final step becomes weaker?

What if companies recover their output without recovering their previous demand for labor?

What if artificial intelligence allows production to rise while employment, hours or wages recover only partially?

Then a mechanism designed to bridge a temporary gap may be asked to support a structural change.

The Fiscal Feedback Loop

The problem does not require every job to disappear. It only requires labor income to grow more slowly than the obligations built around it.

Any one company can benefit from substituting technology for labor.

The company reduces cost. Output rises. Profit may rise. Its customers may receive a better or less expensive product.

But the government does not see only the company's income statement.

It sees the displaced paycheck, the reduced payroll tax, the reduced household spending and the claim for assistance.

This does not mean the company did anything irrational or immoral. Under competitive pressure, it may have had little choice.

It means a decision that is rational at the company level can create a problem at the system level.

Government can borrow to cover the gap, particularly during a temporary downturn. Borrowing is one reason the safety net can keep working when tax receipts fall. But debt can bridge a transition; it cannot permanently replace a shrinking tax base without increasing interest costs or eventually forcing changes in taxes, benefits or both.

But What If AI Makes the Economy Much Richer?

This is the strongest counterargument, and it should not be dismissed.

If artificial intelligence produces a large increase in productivity, the economy may generate far more total income. Lower prices could raise living standards. New industries could appear. Workers who remain employed could earn more. Corporate profits and investment returns could rise enough to generate additional tax revenue.

That outcome is possible.

But a richer economy does not automatically create a solvent safety net.

The question is not only how much income the economy produces.

The question is where that income appears, who receives it and how it is taxed.

A dollar paid as wages enters the payroll tax system immediately. It is also likely to be spent because most households use a large share of their income for current living expenses.

A dollar that appears instead as retained corporate earnings, unrealized asset appreciation or income shifted across jurisdictions does not necessarily support public revenue or consumer demand in the same way or at the same time.

The International Monetary Fund has warned that large labor displacement can erode the tax base as labor's share of income declines. It argues that stronger taxation of capital income may be necessary to finance greater education and social spending in a more automated economy.⁴

More recent IMF scenario work adds an important qualification: shifting taxes from labor to capital may recover only part of the lost revenue because capital is mobile, multinational firms can move income and countries compete to attract investment.⁵

A nation can become more productive while its tax system becomes less aligned with the place where income is being created.

This Is Not Only a Future Mass-Unemployment Problem

We do not need to wait for a world in which most people have no jobs.

The pressure could begin much earlier.

A company hires fewer graduates. A department replaces two departing employees with one. Freelance income becomes less dependable. Full-time positions become part-time positions. Wage growth falls behind productivity growth. Experienced employees use AI to absorb work once performed by junior employees.

None of those changes necessarily produces a dramatic unemployment statistic.

But each can reduce taxable labor income relative to the path governments expected when they made long-term promises.

Social Security already faces demographic pressure because the number of workers supporting each beneficiary has declined over time. AI does not have to cause mass unemployment to add another pressure. It only has to reduce taxable payroll below what it otherwise would have been.

The same principle extends beyond retirement.

State and local governments depend on combinations of income, sales and property taxes. A prolonged weakening of labor income can affect all three: less income to tax, less spending to tax and eventually more strain on household finances and property values.

The consequences would not arrive evenly. A highly automated economy could produce extraordinary wealth in one region or sector while weakening the revenue base in communities where displaced workers actually live.

Can't We Simply Tax the Robots?

“Tax the robots” sounds like an obvious answer.

It is not that simple.

What exactly is a robot?

Is it an industrial arm replacing an assembly worker? Software drafting a legal document? A spreadsheet formula eliminating clerical work? A self-checkout machine? A more efficient piece of equipment that allows one technician to perform the work of two?

A narrowly defined robot tax could encourage companies to redesign technology around the definition. A broad automation tax could penalize beneficial productivity, including tools that make work safer or help workers rather than replace them.

Capital can also move. If one country heavily taxes AI deployment while another does not, investment and reported profits may shift without solving the global problem.

That does not mean the tax system should remain unchanged.

It means the objective should be clearer than the slogan.

The Tax Base May Have to Follow the Income

If income moves away from labor and toward capital, platforms, intellectual property and highly concentrated firms, the tax base may eventually have to move with it.

That could involve some combination of stronger capital-income taxation, broader corporate tax bases, excess-profit taxes, consumption taxes paired with protections for lower-income households, social wealth funds, public ownership stakes or levies connected to the value produced by automated systems.

Each option has tradeoffs.

Tax capital too aggressively in one jurisdiction and it may move. Tax consumption without compensation and the burden can fall disproportionately on people with the least income. Tax automation carelessly and we may discourage tools that improve safety and living standards. Rely entirely on corporate taxes and governments may find themselves bargaining with companies that can relocate income across borders.

There is no single painless replacement for a broad labor-income tax base.

That is precisely why the question should be addressed before displacement becomes large.

Tax systems change slowly. International agreements change even more slowly. Artificial intelligence does not appear to be waiting for either one.

The Safety Net Cannot Be the Entire Economic Model

There is also a deeper limitation.

A safety net is designed to prevent a fall from becoming a catastrophe.

It was not designed to become the permanent economic role of a large share of the population.

If AI substantially reduces the need for human labor, the answer cannot be limited to making unemployment insurance last longer or expanding food assistance.

Those programs may be necessary during a transition. They do not answer the long-term question of how people participate in the income generated by an automated economy.

A society in which a small group owns the productive systems and everyone else repeatedly applies for assistance is not a stable version of abundance.

The durable answer may require new ways for people to share in the productive assets, profits or public revenue created by AI—not merely compensation after their labor income disappears.

The Question Comes Before the Crisis

The debate over artificial intelligence often jumps from one extreme to another.

Either AI will create so much wealth that financing will somehow take care of itself, or AI will eliminate nearly every job and require an immediate universal basic income.

There is a large and consequential territory between those outcomes.

Labor income could decline gradually. Entry-level hiring could weaken first. Payroll growth could trail output. More people could move in and out of unstable work. The need for support could rise long before the tax base collapses.

That is the period in which policy still has room to adapt.

We should use it.

That question is not an argument against artificial intelligence.

It is an argument against building an automated economy on a tax and support system designed for a labor-abundant one.

The safety net cannot be financed indefinitely by assuming the labor income it protects will always remain there to fund it.