AI Tax Policy Considerations | Artificial Intelligence (AI)

Artificial intelligence (AI) is one of the biggest topics in the world—it was only a matter of time until a taxA tax is a mandatory payment or charge collected by local, state, and national governments from individuals or businesses to cover the costs of general government services, goods, and activities. angle materialized. Fears that AI will have a massive effect on the labor market and broader economy have spurred arguments that tax policy should hinder its adoption or address its employment side effects. Sen. Bernie Sanders (I-VT) and Sen. Mark Kelly (D-AZ) have both advocated for various AI taxes, as has Dario Amodei, the CEO of Anthropic, one of the largest AI research labs.

AI technology has a wide range of potential economic outcomes. It would be a mistake to tailor tax policy to the extreme end of that range. Still, policymakers shouldn’t do nothing: they should instead make tax reforms that are sound regardless of which AI scenario emerges.

We Don’t Know What Will Happen

Many bold AI predictions, based on hype or fear, revolve around AI fully replacing human labor. A recent National Bureau of Economic Research workshop explored the economics of such a scenario. While a useful theoretical exercise, we shouldn’t base policy on it.

To date, we do not have evidence of massive-scale disruption. While the job market for entry-level white-collar jobs has slowed recently, the case for AI as the main driver is weak. Analysis from Yale Budget Lab shows signs of transformation in the labor force, but similar to the effects of other disruptive technologies, like personal computers. At the macro level, the labor market is strong, though softening: an unemployment rate of 4.4 percent and a prime-age labor force participation rate of 83.7 percent constitute a stronger labor market than most of the 2010s.

Of course, a scenario where AI is disruptive at a scale incomparable to historical technological advances is still possible. But designing policy now based on the assumption that such a scenario is imminent is misguided.

Against an AI Excise TaxAn excise tax is a tax imposed on a specific good or activity. Excise taxes are commonly levied on cigarettes, alcoholic beverages, soda, gasoline, insurance premiums, amusement activities, and betting, and typically make up a relatively small and volatile portion of state and local and, to a lesser extent, federal tax collections.

The idea of an excise tax to protect workers comes in a few packages. Some specifically target new AI tools, like large language models, while others would apply to automation technologies regardless of type.

Sen. Kelly’s AI for America proposal includes an “AI Horizon Fund,” supported by revenues extracted from the industry. While the proposal does not specify the tax mechanism, it suggests (among other options) taxes on AI industry profits or revenues.

Sen. Sanders’s recent report advocates punitive policies on both physical and digital automation. He proposes eliminating full deductions for equipment investment and adding excise taxes on automating technology. His report cites arguments from economist Daron Acemoglu that the tax code advantages automation.

The AI lab Anthropic organized a symposium of researchers to consider policy responses to AI. Economists Lee Lockwood and Anton Korinek advocated a range of taxes on “token generation, robots, robot services, and digital services.”

These proposals have a few problems.

For starters, the tax code does not advantage automation. Firms can fully deduct worker compensation. That is not always the case for capital investment. Historically, companies must spread deductions for capital investments over several years—meaning they cannot deduct the full real value. The One Big Beautiful Bill Act put investment in short-lived assets at parity with operating costs like wages, but most long-lived assets are still penalized.

Thinking of AI in the same context as other technological innovations shows how foolhardy targeted taxes could be. Imagine excise taxes explicitly designed to slow down railroads in the late 19th century or taxes on inventory management software in the 1980s and 1990s. These would be quite bizarre. And we have practical examples of the deleterious effects of taxing capital inputs: while not intended to slow technological adoption, tariffs on capital goods slowed accumulation and growth in the 1890s.

Slower technology adoption rates drive divergences in economic growth. Recent Federal Reserve research shows that less investment in new technologies helps explain the divergence between European and US GDP per capita since 2000. Economic stagnation in Japan since 1990 has many causes, but the slow adoption of software technology is one of them. And 19th– and 20th-century economic history is full of nations that fell behind due to a reluctance to adopt new technologies.

Explicitly designing policies to slow adoption is therefore a foolish approach.

Reforms to Unemployment Insurance and Reskilling Are Better Alternatives

AI making human labor entirely obsolete is unrealistic. However, AI will disrupt specific occupations just like other periods of technological change. Policy reforms could help smooth the adjustment process for workers by removing tax barriers to hiring and investing in worker training.

Policymakers could make the hiring process more fluid. The US’s unemployment insurance (UI) system relies on experience rating, which means different firms are subject to different tax rates. This design was intended to discourage firing employees, but it ultimately pushes employers to avoid hiring (particularly “risky” employees, such as ones undergoing a career transition) and prefer independent contractors that exist outside the UI system. Reducing or eliminating the role of experience rating in the UI system could make re-entry to or reallocation within the workforce easier.

Policymakers could also encourage reskilling workers. The tax code contains several policies for education and workforce development. However, it caps the deduction for business investment in worker training. A reform package that raised that cap and eliminated ineffective tax credits would make it easier for companies to retrain workers for new roles if technology replaces them in their current ones.

Two Diverging Views on How AI Will Affect Deficits

AI fear and hype have not just led to tax proposals targeting the technology for its potential impact on the labor market. They have also shaped expectations about AI’s impact on the deficit. There are two stories.

The pessimistic story notes that the United States relies on income taxes. It worries AI-driven automation will reduce labor income share and lead to a decline in revenue, forcing a radical redesign of the tax system. The offered solution is to target AI companies with specific new taxes—not to reduce AI adoption, just to replace lost income tax revenue.

The optimistic story, told by Elon Musk and others, says AI will drive rapid productivity growth, which will raise revenue and accordingly reduce deficits. This deficit reduction will therefore reduce the need for spending cuts or tax increases to rein in unsustainable structural deficits.

Which story is right?

If we assume AI will contribute positively to productivity and economic growth, a prerequisite for disruption concerns, we should expect it to improve the US fiscal situation. The labor share of net income (subtracting out depreciationDepreciation is a measurement of the “useful life” of a business asset, such as machinery or a factory, to determine the multiyear period over which the cost of that asset can be deducted from taxable income. Instead of allowing businesses to deduct the cost of investments immediately (i.e., full expensing), depreciation requires deductions to be taken over time, reducing their value and disco and production taxes) has remained around 70 percent since 1940. If incomes rise, tax revenue should rise. The Congressional Budget Office (CBO) finds faster productivity growth will raise revenue and reduce deficits, as would Tax Foundation’s model.

If AI adds to productivity, it should reduce deficits. However, AI adding to productivity growth does not mean an improvement over CBO’s projections. CBO’s baseline already includes annual productivity growth of between 1.5 and 1.7 percent. While AI may contribute to productivity growth, it might only meet the baseline. That would mean no alteration to the fiscal picture—not fixing our deficit problems but not making them worse either.

Shifting to Consumption Taxes Is Worthwhile Either Way

It is not likely that AI-driven productivity growth will increase the deficit, but some tax reforms, like shifting to consumption taxes, would still be worth pursuing.

The United States is the only developed country without a value-added tax (VAT). Introducing a VAT to reduce the deficit and replace part of the income tax would be a sound, pro-growth tax reform. Consumption taxes are more stable and less economically damaging revenue sources than income taxes.

Regarding AI profits, we already have a tax that covers them: the corporate income taxA corporate income tax (CIT) is levied by federal and state governments on business profits. Many companies are not subject to the CIT because they are taxed as pass-through businesses, with income reportable under the individual income tax.. And policymakers can make the corporate tax focus on “supernormal” profits by allowing full deductions for both capital and labor costs.

Conclusion

Maybe AI technology is so transformative that it forces a reconsideration of public finance principles, but to date, such a scenario is restricted to the realm of science fiction. And science fiction is not a sound basis for tax reform.