Research Paper
Mapping Tax Risks From Labour-Displacing AI
Trish Ieong, Akbar Saputra, Anuja Maniar and Deric Cheng
This paper addresses an emerging fiscal risk for tax policymakers: the potential for powerful, productivity-enhancing AI systems to erode tax bases through labour displacement. We identify four channels through which fiscal pressures may arise, simulate scenarios for an ‘average OECD country’ under various assumptions, and examine country-specific factors likely to produce different results across jurisdictions. Our aim is to prompt governments to start preparing now, and we offer some high-level suggestions for doing so.
Published April 24th 2026
Executive Summary
This paper addresses an emerging fiscal risk for tax policymakers: the potential for powerful, productivity-enhancing AI systems to erode tax bases through labour displacement. We identify four channels through which fiscal pressures may arise, simulate scenarios for an ‘average OECD country’ under various assumptions, and examine country-specific factors likely to produce different results across jurisdictions. Our aim is to prompt governments to start preparing now, and we offer some high-level suggestions for doing so.
Artificial intelligence (AI) has the potential to displace workers at a scale and pace that could create serious fiscal challenges for governments. Even if AI increases overall productivity, fiscal pressures could arise because of the tax structure of most OECD countries.
Part One discusses how these pressures could arise through four distinct channels:
First, labour income is typically subject to higher effective tax rates than capital income, such that when AI displaces a worker, income shifts from a highly-taxed base to a lower-taxed one.
Second, where the labour income of a country’s resident workers is replaced by capital income accruing to non-resident providers of AI, countries would face additional tax leakage as they have limited taxing rights over non-residents’ income.
Third, some productivity gains may flow to consumers in the form of lower prices or free services, creating untaxable consumer surpluses rather than taxable income.
Fourth, governments may face increased spending pressures to support displaced workers through income support, retraining, and other safety net provisions.
In Part Two, we present a scenario-based fiscal simulation that illustrates the potential magnitude of these fiscal pressures for an ‘average OECD country’. The simulation uses an accounting-based framework based on two key parameters: the degree of labour market disruption (low, medium, or high), which determines how many workers are displaced, and the level of domestic value capture (high or low), which determines how much of AI’s productivity gains remain within the domestic tax base.
The simulation is not intended to act as a forecast, but rather to help policymakers understand the mechanisms and the range of outcomes produced under different assumptions.
The following table summarises the cumulative 10-year changes under six different scenarios in the simulation.
Scenario
Total Tax Revenue Change
Net Fiscal Impact Change
Low impact (12% displacement);
High domestic value capture (90%)
0.2%
-0.8%
Medium impact (30% displacement);
High domestic value capture (90%)
-2.8%
-8.4%
High impact (48% displacement);
High domestic value capture (90%)
-7.3%
-20.0%
Low impact (12% displacement);
Low domestic value capture (30%)
-1.8%
-2.8%
Medium impact (30% displacement);
Low domestic value capture (30%)
-7.7%
-13.4%
High impact (48% displacement);
Low domestic value capture (30%)
-15.3%
-28.0%
In Part Three, we examine key uncertainties and simplifications of our model. Some of these simplifications are likely to result in overestimation of fiscal pressures; others are likely to result in underestimation. We discuss three key uncertainties in detail: how comparable AI is to past episodes of labour displacement; the degree of potential output expansion; and the effect on non-displaced workers’ wages. The net directional effect of these uncertainties is unclear.
While our estimates are for an ‘average OECD country’, actual impacts are likely to vary considerably across countries. Part Four discusses the factors likely to drive such variation: the gap between effective tax rates on labour and capital income; the concentration of labour income; overall tax base composition; exposure to AI; how AI affects terms of trade; whether the country is a net capital exporter or importer; demographics; and existing fiscal headroom.
In Part Five, we offer some high-level suggestions for policymakers. Because there is considerable variation across countries and tax systems, we do not prescribe specific reforms. Instead, we recommend that policymakers actively monitor developments in AI, and coordinate with other agencies. Policymakers should also stress-test their own tax systems, potentially planning for broader, structural reforms where those tests reveal vulnerabilities.
We urge policymakers to treat the possibility of mass AI-driven labour displacement as a long-term fiscal risk on par with, if not exceeding, the challenges of an ageing population. While demographic shifts unfold over generations, the pace of AI progress is unpredictable and breakthroughs can be sudden. Governments that take steps now to broaden their tax bases will be better positioned to benefit from AI’s productivity gains, while those that ignore the problem may find themselves left with only the most drastic options.
