SSRN Review & Roundup: Eyal Reviews Harpaz’s Taxing AI

This week, Mirit Eyal (Alabama; Google Scholar) reviews a new work by Assaf Harpaz (Georgia; Google Scholar), Taxing AI, B.U. L. Rev. __ (forthcoming 2026):

Artificial intelligence is no longer a distant prospect confined to science fiction or academic speculation. It is now reshaping industries, displacing workers, and concentrating wealth at an unprecedented pace. The federal tax system, heavily reliant on taxing individual labor income and payroll, finds itself particularly vulnerable to these shifts. AI-driven efficiency gains led to over 55,000 U.S. job cuts in 2025, a dramatic increase from previous years as companies in tech, finance, and customer service pivot to automation. Major firms including Amazon, Microsoft, and Google have reduced headcount or paused hiring, citing the need to lean into generative AI tools for coding, marketing, and administrative tasks. Against this backdrop, Harpaz’s Article offers a timely and rigorous examination of how AI threatens to disrupt the tax system’s fundamental goals and what can be done about it.

Harpaz opens with a striking historical parallel. In 1984, when a Pennsylvania Volkswagen plant announced it would replace hundreds of workers with robots, a constituent proposed that robots be required to pay federal income taxes. The Treasury responded that inanimate objects do not file returns, and that robots “figuratively and literally pay income tax” through the incomes of workers and owners. Forty-one years later, the question has returned with renewed urgency, as companies like Mechanize openly pursue the goal of automating the entire labor market. Harpaz uses this framing to underscore a central tension: while the doctrinal answer remains the same, the economic and distributive realities have changed dramatically.

The Article’s methodological contribution lies in its functionalist approach. Rather than asking whether AI challenges the plain language of the Internal Revenue Code, which it largely does not, Harpaz asks whether AI undermines the purposes the Code is meant to serve. He identifies three core functions of the federal tax system: revenue-raising, redistribution, and regulation of taxpayer behavior. Each of these functions, he argues, is threatened by AI’s capacity to shift economic value from labor to capital, concentrate wealth among a small number of firms and investors, and erode the tax base that funds government programs.

Harpaz’s analysis of revenue loss is particularly compelling. He notes that approximately half of federal revenues derive from individual income taxes and a third from payroll taxes, both tied to human labor. If AI displaces workers-at-scale, these revenue streams will contract. He illustrates this with a thought experiment: If Mechanize succeeds in automating all U.S. labor, the federal government could no longer collect revenues from individual labor income or fund Social Security and Medicare. While this scenario is extreme, it dramatizes the structural vulnerability of a tax system that has no broad-based consumption tax and treats capital gains preferentially.

The distributive dimension of Harpaz’s argument is equally forceful. He documents the extraordinary concentration of AI-related wealth among Big Tech companies and a handful of startups. Nvidia’s rise to a $5 trillion market capitalization, powered by the AI frenzy, exemplifies how AI gains accrue to a narrow slice of firms and investors. Lower-wage, less-educated workers face displacement without the mobility or resources to transition to new sectors. Meanwhile, the tax system continues to favor capital over labor through preferential rates on long-term capital gains and qualified dividends, accelerated depreciation, and bonus expensing. Harpaz persuasively argues that this asymmetry exacerbates the distributive harms AI is already generating.

The Article’s engagement with existing proposals is thorough and balanced. Harpaz examines “robot taxes”, namely various levies on firms that automate, and finds them conceptually flawed. Robot taxes treat AI as a negative externality akin to pollution, when in fact technological progress should be encouraged rather than penalized. They also face significant definitional and administrative challenges: what counts as a “robot,” and how would imputed salaries be calculated? Moreover, unilateral adoption of robot taxes could disadvantage domestic industries by incentivizing firms to relocate production abroad.

Harpaz is equally skeptical of proposals to assign tax personhood to AI systems. While scholars have argued that autonomous AI models could be treated as taxable entities, Harpaz notes that AI currently lacks legal personhood, and the ultimate tax burden would still fall on natural persons. Creating AI tax personhood would require statutory reform, a new doctrinal framework, and answers to difficult questions about which AI systems qualify and what tax regime applies. These proposals, while intellectually interesting, remain, in Harpaz’s opinion, impractical in the near term.

The Article’s scholarly contribution lies in its two proposed interventions. In the short term, Harpaz recommends increasing capital gains rates on the sale of ownership interests in AI-intensive firms. Drawing on existing models like the net investment income tax and the collectibles gain regime, he proposes a surtax that would reduce or eliminate the preferential treatment currently afforded to AI-related capital gains. This approach targets the specific wealth concentration problem AI creates without undermining the broader tax preference for capital. Harpaz acknowledges the design challenges, particularly how to define “AI-intensive” firms but argues that Congress has successfully used categorical tests in analogous contexts, such as the passive foreign investment company regime.

In the long term, Harpaz suggests adopting a broad-based consumption tax if labor income’s share of the tax base materially declines. The United States remains an outlier among developed economies in lacking a value-added tax. While consumption taxes are often viewed as regressive, Harpaz notes that the revenues they generate can fund progressive programs, including measures to mitigate AI’s distributive harms. He contrasts this approach with targeted excise taxes, which may function as Pigouvian levies and face the same conceptual objections as robot taxes.

There are several areas where the Article could be strengthened. First, while Harpaz’s functionalist methodology is analytically powerful, it could benefit from deeper engagement with the political economy of tax reform. He acknowledges that capital favoritism is “inseparable from American capitalism” and that wealthy constituencies exert disproportionate influence over lawmakers, but these observations remain somewhat underdeveloped. A more sustained analysis of why the tax system has proven resistant to reform (but would work for the reform proposals he suggests), and how AI-specific proposals might overcome that resistance, would enhance the Article’s practical impact.

Second, the proposed surtax on AI-intensive firms raises implementation questions that deserve fuller treatment. Harpaz suggests that firms could be classified based on AI-related research and development expenditures, revenue from AI products, or the proportion of value attributable to AI intangibles. These metrics are plausible, but firms have strong incentives to restructure or reclassify activities to avoid higher rates. A discussion of anti-avoidance measures, safe harbors, or regulatory oversight mechanisms would strengthen the proposal’s credibility.

Third, the Article’s treatment of consumption taxation, while thoughtful, could engage more directly with the distributional design choices a broad-based consumption tax would entail. Harpaz notes that revenues can fund progressive programs, but the specific mechanisms, such as exemptions for necessities, refundable credits, or demogrants, warrant closer examination. The political and administrative feasibility of implementing a federal VAT in the United States also deserves attention, particularly given the historical resistance to such measures.

Finally, the Article might benefit from a more explicit engagement with the global dimensions of AI taxation. Harpaz notes that unilateral robot taxes could disadvantage domestic industries, but the international coordination challenges extend to his own proposals. If the United States imposes a surtax on AI-related capital gains, could investors shift ownership to jurisdictions with more favorable treatment? How would the proposal interact with the OECD’s Pillar Two global minimum tax framework? These questions are not fatal to Harpaz’s argument but addressing them would demonstrate the proposal’s robustness in an interconnected global economy.

Despite these areas for further development, Harpaz’s Article represents a significant contribution to the emerging literature on AI and taxation. His functionalist reframing of the debate, focusing on the tax system’s purposes rather than its plain language, offers a productive analytical lens for evaluating policy responses. His critique of robot taxes and AI personhood proposals is rigorous and persuasive. And his two-track intervention, a short-term surtax on AI capital gains and a long-term shift toward consumption taxation, provides concrete, actionable recommendations grounded in existing statutory models.

This Article arrives at a critical moment. As AI continues to transform the economy, the federal tax system faces an inflection point. Policymakers can either adapt the system to preserve its revenue-raising, redistributive, and regulatory functions, or watch those functions erode under the weight of technological change. Harpaz’s work provides the intellectual foundation for the former path. It helps to develop the discussion among tax scholars, policymakers, and anyone concerned with the fiscal and distributive implications of the AI transformation.

While Harpaz dismisses AI tax personhood as impractical in the near term, David Elkins and I have argued here that such personhood is both conceptually coherent and administratively necessary in carefully defined contexts. Under three scenarios of information opacity, lack of access to AI-controlled assets, and beneficiary indeterminacy in which traditional attribution rules break down, we proposed a flexible tax regime modeled on existing frameworks for trusts and corporations. Read together, the two articles illuminate complementary dimensions of the AI taxation challenge: Harpaz focuses on how AI threatens tax system functions and proposes incremental reforms within the existing framework, while our article Taxing AI Personhood confronts the deeper doctrinal question of when AI itself must be recognized as a taxable entity under administrative realities. The ongoing conversation between these perspectives will be essential as policymakers grapple with the full scope of AI’s implications for income taxation.

Here is the rest of this week’s SSRN Tax Roundup:

Ankur Agarwal (affiliation not provided to SSRN), A Proposal to Address Key Tax Treaty Issues for Non-Resident Entertainers and Sportspersons, 79 Bulletin for Int’l Tax’n, issue 11 (2025)

Benjamin Alarie (Toronto), Cognitive Infrastructure for Public Revenue Systems (Mar. 2026)

Laura Boyd, Thomas Conefrey, Ronan Hickey, Matija Lozej, Boryana Madzharova, Niall McInerney, & Graeme Walsh (all from Central Bank of Ireland), Managing Risks and Building Resilience in the Public Finances (Mar. 2026)

Vaibhav Chaudhari (Independent), A Decade of Corporate Taxation: Empirical Evidence from BSE Sensex 30 Companies (FY2016–FY2025) (Mar. 2026)

Rita De La Feria (Leeds), From “Flatulence Taxes” to “Condom Taxes”: Excises as the New Frontier for EU Tax Law (Mar. 2026)

Jacob Goldin (Chicago) & Zachary D. Liscow (Yale), When Should the Legal System Help Redistribute Income? (Mar. 2026)

Marilyn Hajj (Virginia), Rate Your Experience: Comparative Lessons for Gig Work (Mar. 2026)

Leni Hirsch (affiliation not provided to SSRN), The Tax Treatment of College Sports in the NIL Era: A Case for Reform (Mar. 2026)

Domenico Imparato (Hamburg), The “Hidden Sides” of the Taxation of Dividends and Interest for Corporate America Versus Corporate Europe, in Taxing Corporates (Craig West ed., forthcoming 2026)

Andreas Kallergis (Inst. U. de France), Democratic Control over Tax Treaties: A Comparative Constitutional Perspective (Mar. 2026)

Hanan Katheeb (PES U.), The Comparative Study on Various Taxes Imposed in US State of Alaska (Mar. 2026)