Papers
Topics
Authors
Recent
Search
2000 character limit reached

Taxing Artificial Intelligence

Published 2 Jul 2026 in cs.CY | (2607.02144v1)

Abstract: While AI promises major benefits, its development and deployment can shift costs onto others, including environmental pressures on local communities, labor and creative displacement, and systemic risks from rapid frontier development. Taxation is an integral part of policy design, and recent academic, industry, and policy debates have begun to consider whether tax instruments can help address these harms. In this paper, we explore the viability of AI taxation. More broadly, AI taxation should not be understood only as Pigouvian correction. In the AI context, taxation can also correct harmful activity, redistribute unevenly borne costs and gains, and fund regulatory capacity. We discuss the main externalities associated with AI and survey possible tax instruments, including corporate income and rent-based taxes, consumption taxes on AI-related services, and excise instruments tied to specific AI activities. We further assess the benefits and pitfalls of these instruments, including feasibility, measurement problems, incidence, leakage, and innovation costs. Because AI externalities differ in nuanced ways, tax policy must be carefully designed and matched to the specific harms and policy objectives.

Authors (2)

Summary

  • The paper's main contribution is a systematic framework for applying taxation to AI-induced externalities to correct market failures.
  • It evaluates diverse tax instruments—such as excise, consumption, payroll, and corporate taxes—and their implications for regulating the AI supply chain.
  • The analysis highlights measurement challenges, risks of industry capture, and the need for adaptive policy measures in evolving AI governance.

Taxing Artificial Intelligence: An Analytical Essay

Introduction

The paper "Taxing Artificial Intelligence" (2607.02144) presents a systematic examination of the prospects, justifications, and complexities of implementing taxation as a policy tool for addressing AI-induced externalities. The work moves beyond earlier abstract discussions of "robot taxes" to engage with active legislative proposals and provides rigorous classification and assessment of both the harms caused by AI and the instruments by which taxation could internalize or redistribute these costs.

Theoretical Framework and Motivation

The conceptual foundation rests on the realignment of taxation as not merely a Pigouvian corrective for negative externalities but as a broadly applicable instrument for (i) shaping incentives, (ii) funding regulatory capacity, and (iii) redistribution. The authors identify the defining feature of AI externalities: the decoupling of private gains from diffuse, societal costs, which are often borne by actors with minimal agency in AI's development or deployment. Special attention is given to the adaptability of taxation vis-à-vis other regulatory modalities, exploiting existing fiscal infrastructure and compliance processes.

Characterization of AI Externalities

A central contribution of the paper is its mapping of the AI-induced negative externalities:

  • Resource Consumption: AI-driven data center expansion stresses local energy and water infrastructure, leading to increased utility costs and environmental burdens, which are not internalized by AI actors. The measurement and attribution of AI-related resource use present non-trivial challenges, especially in distinguishing AI activity within generalized data center operations.
  • Labor and Creative Displacement: AI-enabled automation decreases demand for human labor across various segments, including creative work, reducing wage income, eroding the payroll tax base, and threatening the viability of sectors that directly contribute to training data for generative models. The diffusion of labor displacement externalities complicates direct compensation or correction.
  • Environmental, Informational, and Societal Harms: The authors also address emissions, e-waste, misinformation, bias, privacy breaches, cybersecurity threats, and catastrophic risk from frontier AI as underpriced costs manifesting throughout the AI value chain. These often exhibit non-local, temporally diffuse, or contextually contingent characteristics that undermine the efficacy of both direct regulation and straightforward price-based correction.

These externalities are systematically assessed in terms of economic rationale, repercussions, and measurement difficulties. The distinction between externalities amenable to taxation and those where pricing is either infeasible or normatively questionable is well-articulated.

Design and Implementation of AI Taxation

The paper provides an exhaustive taxonomy of tax instruments with respect to AI, discussing key design dimensions:

  • Base Selection: A clear articulation of when to tax observable proxies (data center electricity, token/API usage) versus less tangible quantities (AI-enabled profits/rents) is provided. The choice of tax base operationalizes the regulatory objective—whether correction, redistribution, or regulatory funding.
  • Tax Instruments: The survey includes excise taxes (resource use), consumption taxes (service/API use), payroll tax extensions, corporate/rent/excess-profit taxes, and benchmark-based exemptions for safety or fairness alignment. The legal and practical implications of each instrument, the locus along the AI supply chain, and the likely behavioral responses are precisely characterized.
  • Incidence and Incentives: Emphasis is placed on tax incidence—distinguishing formal payers from ultimate economic bearers—and on minimizing regulatory arbitrage, jurisdiction shopping, and strategic activity reclassification.
  • Implementation and Measurement Challenges: The text highlights the brittleness of legal/technical definitions of "AI," the fragmentary structure of AI supply chains, and the risks of tax-induced distortions or leakage, where activity simply migrates rather than diminishes. The authors stress the importance of instrument selection that leverages existing reporting and audit mechanisms to mitigate these issues.

Crucially, the paper offers strong claims regarding the feasibility and institutional advantages of AI taxation: its measurability (in selected bases), compatibility with existing administrative infrastructure, and relative ease of legislative passage compared to novel regulatory regimes.

Policy Implications and Open Problems

The strategic use of taxation in AI governance is situated as a complementary, not substitutive, tool within the broader spectrum of regulatory responses. AI taxes can rectify incentive misalignments, recycle economic rents toward regulatory or redistributive objectives, and shore up public oversight capacity. The authors, however, foreground substantial challenges:

  • Innovation and Global Competitiveness: Ill-calibrated taxes may undermine domestic AI R&D and induce tax base erosion via relocation or vertical integration.
  • Design and Political Economy Risks: The susceptibility of tax design to industry capture, loopholes, and misdirected incidence is critically assessed. The authors warn that strong industry lobbying could hollow out the tax base or shift liability onto unintended actors.
  • Measurement and Double Counting: Attribution of harms, especially environmental ones, requires caution to avoid either underpricing or double taxation when multiple instruments (e.g., green taxes) co-exist.
  • Corrective Limits: For some externalities (notably rights-based harms like privacy and systemic discrimination), pricing may be both empirically and normatively indefensible.
  • Dynamic Adaptation: The necessity for tax design iteration, adaptive thresholds, and potential international coordination is strongly underscored.

Future Directions

The authors indicate that the evolution of AI governance will likely demand hybridized mechanisms, where taxation interfaces with sectoral regulation, standards, and auditing regimes. International agreements on digital/AI taxation and real-time calibration of bases and rates in response to economic, technological, and social feedbacks are anticipated as critical research and policy frontiers.

Conclusion

"Taxing Artificial Intelligence" supplies a robust analytical foundation for AI fiscal policy, articulating both the economic logic and limitations of taxation as a response to AI externalities. The essay demonstrates that whereas taxation does not yield a panacea for all AI-induced harms, targeted, well-calibrated instruments can internalize or redistribute costs, support regulatory adaptation, and ameliorate some deficiencies of existing law. The multi-instrument, context-sensitive approach outlined suggests a developmental trajectory for AI governance that privileges institutional realism and adaptability over prescriptive or monolithic solutions.

The careful design of AI tax instruments, particularly their base, incidence, and supply-chain positioning, alongside coordinated regulatory oversight, emerges as a central pillar for any future-facing AI policy architecture.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Sign Up to Generate All Videos Subscribe on YouTube

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Sign Up to Generate

Explain it Like I'm 14

A simple guide to “Taxing Artificial Intelligence”

Overview: What is this paper about?

This paper asks a big, practical question: Can taxes help deal with the downsides of AI? AI can bring great benefits, but it can also push costs onto other people—like higher local water and electricity bills, job losses for some workers and artists, or risks from very powerful systems. The authors argue that, if designed carefully, taxes can:

  • discourage harmful AI-related behavior,
  • share AI’s gains more fairly,
  • and fund the government teams needed to oversee AI safely.

Key goals and questions

The paper looks at five easy-to-understand questions:

  1. What kinds of harms does AI cause for people who didn’t choose them or get paid for them (called “externalities”)?
  2. Why might we tax AI—what purposes can a tax serve?
  3. How do taxes fit with other AI rules and laws?
  4. How could an AI tax be designed in practice?
  5. What can go wrong with AI taxes, and how can we avoid that?

Approach: How did the authors study this?

This is a policy and economics analysis, not a lab experiment. The authors:

  • Review real-world examples, like data centers using lots of water and energy, and generative AI affecting artists’ income.
  • Summarize laws and proposals being discussed by governments and industry leaders.
  • Explain common tax tools (like income taxes, sales/consumption taxes, and excise taxes) and how they might apply to AI.
  • Lay out a step-by-step framework for designing an AI tax and warn about pitfalls.

Helpful translations of key terms:

  • Externality: When someone’s actions cause costs (or benefits) to others that aren’t included in the price. Example: a factory pollutes, and neighbors breathe the smoke. They pay the cost, not the factory.
  • Pigouvian tax: A “pay for the mess you make” tax. It puts a price on harmful activity to discourage it (like a pollution tax).
  • Tax base: What you measure and tax (like gallons of water used, kilowatt-hours of electricity, or profits).
  • Taxpayer vs. remitter: The taxpayer is who owes the tax by law. The remitter is who actually sends the money to the government. Sometimes they’re the same, sometimes not (for example, stores remit sales tax collected from customers).
  • Tax incidence: Who ends up bearing the cost after prices change. A company might raise its prices, so customers pay part of the tax.
  • Excise tax: A tax on a specific thing or activity (like a gasoline tax per gallon).

Main findings: What did they learn?

  1. Three strong reasons to tax AI
    • Correct harmful activity: By putting a price on specific AI-related behaviors (for example, heavy electricity or water use), taxes can nudge companies to do less of the harmful thing or do it more efficiently.
    • Redistribute gains and costs: AI profits often go to a few firms, while costs hit ordinary people (nearby residents, artists, workers). A tax can raise money to help those affected—for example, pay for worker retraining or lower local utility bills.
    • Fund oversight: Good AI oversight needs money and skilled people. Tax revenue can build this capacity.
  2. Different harms need different tax designs The paper lists multiple potential AI externalities—higher local energy and water prices, labor and creative displacement, misinformation, bias, privacy risks, cybersecurity problems, and even catastrophic risks. Because these harms are so different, a one-size-fits-all tax won’t work. You have to match the tax to the harm and the goal.
  3. Real-world examples
    • Data centers and utilities: AI needs a lot of computing power, which uses electricity and often water for cooling. This can raise local bills and strain infrastructure. Possible tax responses include:
      • an excise tax on electricity or water used,
      • a tax on “reserved capacity” (to discourage unpredictable, spiky demand),
      • and using the revenue to upgrade the grid, fund recycled water systems, or offset residents’ bills.
    • Creative work and generative AI: AI models are trained on huge amounts of human-created content. Creators may not be asked or paid, and AI outputs can reduce demand for some human work. Possible tax responses include:
      • taxes on profits from generative AI products,
      • taxes tied to training practices (for example, how much non-synthetic data is used),
      • and credits or exemptions for firms that pay creators or disclose their data sources. Revenue could fund creator support, licensing systems, or new tools for tracking data use.
  4. What kinds of taxes could be used? The paper surveys several options:
    • Excise taxes on measurable activities (electricity, water, compute, API usage).
    • Consumption taxes (like sales tax) on AI services.
    • Corporate income or “rent”/windfall taxes when firms earn unusually high profits from AI.
    • Payroll or related taxes if the goal is to address job displacement or support workers.
  5. Practical advantages of using taxes
    • Governments already know how to run tax systems and audit them. That means faster rollout.
    • Companies already file taxes; this can be less burdensome than building a brand-new regulatory system.
    • In the U.S., tax bills can sometimes pass with a simple majority using “budget reconciliation.”
  6. Big challenges and risks
    • Measuring and targeting: It can be hard to separate AI-related electricity use from non-AI use, or to link a tax directly to harms like misinformation.
    • Definitions and gaming: “What counts as AI?” is blurry. Firms might reclassify activities or move operations to avoid taxes.
    • Incidence and fairness: Companies might pass the costs to customers, small businesses, or developers using their APIs.
    • Innovation and competitiveness: Poorly designed taxes could slow useful AI progress or push it overseas.
    • Political capture: Powerful firms might lobby for loopholes.
    • Not all harms fit taxation: Rights-based harms (like privacy or discrimination) may need direct rules, not just taxes.

Why this matters: Possible impact

If used wisely, AI taxes can be a practical tool to make AI development fairer and safer. They can:

  • push companies toward more efficient and responsible practices,
  • raise money to protect and support the people most affected by AI,
  • and finance strong oversight so other AI laws actually work.

But taxes are not a magic fix. They should sit alongside other tools—like transparency rules, safety audits, incident reporting, and clear standards for fairness and privacy. The main lesson is to match the tax to the specific harm and goal, measure what’s feasible, keep loopholes small, and use the revenue to build a safer, more balanced AI future.

View Paper Prompt  View All Prompts 

Knowledge Gaps

Below is a focused list of concrete knowledge gaps, limitations, and open questions the paper leaves unresolved, structured to guide future research and policy design.

  • Lack of empirical estimates of marginal external costs for specific AI harms (e.g., per-kWh/per-gallon local price impacts, misinformation harms, bias/discrimination costs, systemic/catastrophic risk), which are needed to calibrate Pigouvian tax rates.
  • No operational protocol for attributing data-center electricity and water use to AI workloads versus non-AI workloads in multi-tenant cloud/data center environments.
  • Absence of auditable, tamper-resistant measurement methods for prospective tax bases such as compute (FLOP/s, GPU-hours), tokens, or model-training intensity.
  • Unsettled criteria for selecting the taxpayer along the AI supply chain (chipmakers, cloud providers, model developers, finetuners, API intermediaries, deployers) for each externality and objective.
  • Insufficient incidence analysis quantifying who ultimately bears different AI taxes (large firms vs SMEs, consumers vs enterprise customers, creators vs platforms), by income, industry, and geography.
  • No framework for coordinating AI tax policy across jurisdictions to mitigate leakage/offshoring (state-to-state in the U.S., or internationally), including nexus rules for cloud services and cross-border inference.
  • Unclear interactions with existing taxes, subsidies, and tariffs (e.g., R&D credits, accelerated depreciation, electricity tariffs, carbon pricing, water rights), and the risk of double-counting or countervailing incentives.
  • Lack of concrete anti-avoidance and anti-gaming rules for thresholds (e.g., compute caps, model size cutoffs) and for activity reclassification (labeling “AI” as generic software or moving training offshore).
  • No enforceable, technology-neutral definition of “AI activity” suitable for tax law that minimizes ambiguity and circumvention.
  • No validated proxies that correlate with systemic/catastrophic risk to support risk-sensitive taxation (e.g., training compute thresholds, red-teaming outcomes, model autonomy metrics).
  • Insufficient guidance on when rights-based harms (privacy, discrimination) are inappropriate for taxation and which non-fiscal tools should be prioritized and funded instead.
  • Administrative capacity gap: how tax authorities will build/retain technical expertise and tooling (e.g., telemetry, audits, compute attestation) to administer and enforce AI-specific taxes.
  • Missing enforcement architecture for AI taxes (penalties, audit frequency, third-party attestation, cryptographic/hardware-based measurement, role of utilities as verifiers/remitters).
  • No dynamic adjustment rules (indexation) for tax rates as model/data efficiency, cooling technology, and grid/water scarcity evolve; lack of sunset/review triggers.
  • Revenue earmarking and governance remain unspecified: how to structure AI funds (regulatory capacity, redistribution), ensure independence, prevent capture, and set transparent allocation rules.
  • Lack of distributional design tools (progressivity, small-firm exemptions, de minimis thresholds) to avoid entrenching incumbents or burdening startups and open-source communities.
  • No comparative cost-effectiveness analysis of taxation versus alternative instruments (liability regimes, permits, compute caps, cap-and-trade for emissions/compute, direct standards).
  • Unclear design and verification for creator-compensation-linked credits: how to measure “sufficient” compensation, prevent “license washing,” and audit provenance/disclosure claims.
  • Treatment of synthetic data is unspecified: whether taxes/credits should distinguish synthetic vs human-sourced data and how to avoid perverse incentives toward low-quality or self-referential data.
  • No allocation methodology for apportioning shared data-center resource use (energy, water) to individual AI tenants/jobs with auditability (workload-level metering, scheduler logs).
  • Open question on how to tax reserved capacity versus realized consumption to correct demand-uncertainty externalities without incentivizing over-contracting or relocation.
  • No proposed metrics/proxies for taxing misinformation externalities that avoid constitutional/speech risks while targeting concrete, measurable harms (e.g., fraud, clear safety violations).
  • Cybersecurity externalities are acknowledged, but no tax-compatible security metrics (e.g., verified secure development lifecycle, vulnerability disclosure performance) are proposed for credits/penalties.
  • Budget-reconciliation constraints (e.g., U.S. Byrd Rule) are noted but not translated into concrete drafting guidelines that keep AI-tax provisions germane and scoreable.
  • No international template for AI sector taxation analogous to OECD Pillar Two, including minimum taxes, allocation keys (e.g., user location, data-center location), and dispute resolution.
  • Absent phase-in/transition strategies (pilots, sandboxes, graduated rates) to limit shocks and learn before scaling AI tax regimes.
  • Potential perverse effects of proxy bases are not quantified (e.g., taxing tokens may push toward token-efficient but more compute-intensive models; taxing electricity may increase water use).
  • Lack of high-resolution, public datasets linking data centers to local water/energy prices, reliability, and infrastructure costs, necessary for targeting and evaluating taxes.
  • No detailed plan for ensuring redistributive revenues reach affected communities (residents near data centers, displaced workers, creators) with low administrative burden and measurable outcomes.
  • Worker displacement channels are not quantified by sector/region, limiting targeted tax design and evaluation of retraining/upskilling uses of revenue.
  • Cross-border creator compensation is unaddressed: how to identify eligible creators internationally and apportion funds from globally trained models.
  • Governance and audit standards for “AI regulatory capacity” funds are unspecified (eligibility of spending on audits, evaluations, incident tracking; performance metrics).
  • Federalism/preemption questions are open: how state-level AI taxes interact with federal policy, potential preemption, and how to manage interstate tax competition.
  • Revenue volatility is not addressed: how to design stable funding streams for oversight given cyclicality in AI investment and usage.
  • Shadow/hidden compute risk is not addressed: detection and enforcement against undeclared training clusters, multi-entity structures, and beneficial ownership obfuscation.
  • Privacy considerations in usage-based taxes (e.g., API-level logs to calculate tax) are not resolved; need minimization strategies and legal safeguards.
  • Verification for model-level triggers (e.g., total training FLOPs) is unspecified: the role of hardware attestation, secure enclaves, logs, or independent monitors to prevent underreporting.
  • Interactions with carbon pricing are unmodeled: whether AI-specific energy taxes double-count emissions costs and how to coordinate with carbon markets.
  • Multi-objective design is not developed: how to jointly tax electricity, water, and carbon to avoid burden-shifting across environmental resources.
  • Evaluation methodology is missing: experimental/observational designs to assess enacted taxes (e.g., Virginia electricity tax, Ohio tariffs) on behavior, prices, environmental and welfare outcomes.
View Paper Prompt  View All Prompts 

Practical Applications

Immediate Applications

The following applications can be deployed with today’s tax, utility, and compliance infrastructure, drawing on the paper’s proposed instruments (excise, consumption, corporate/rent), stated objectives (corrective, redistribution, regulatory capacity), and case studies (data centers, creative labor).

  • Bold: Local data‑center electricity excise with community earmarks
    • What: Per‑kWh excise on data-center AI loads to price local grid impacts; revenue funds grid upgrades, resident bill rebates, and environmental mitigation.
    • Sectors: Energy, Cloud/Datacenters, State/Local Policy, Utilities.
    • Tools/Workflows: Utility metering and remittance; community benefit dashboards; peak‑aware procurement and workload scheduling; third‑party audits.
    • Assumptions/Dependencies: Legal authority for excise; ability to identify eligible facilities; risk of cost pass‑through and offshoring; state PUC coordination.
  • Bold: Water‑use charges and recycled‑water credits for AI cooling
    • What: Per‑gallon excise on potable water used for AI cooling; credits or lower rates for reclaimed/gray water use.
    • Sectors: Water/Environment, Cloud/Datacenters, State/Local Policy.
    • Tools/Workflows: Water utility billing integration; certification for recycled water systems; site‑selection criteria updates.
    • Assumptions/Dependencies: Metered usage and verification; availability of reclaimed water; potential incidence on residents if firms pass costs on.
  • Bold: Reserved‑capacity (demand uncertainty) surcharge
    • What: Minimum‑bill or reservation‑based tax on contracted grid capacity to internalize planning costs from volatile AI loads.
    • Sectors: Energy, Utilities, Policy.
    • Tools/Workflows: Tariff‑style schedules administered via utilities; predictable compute procurement contracts; grid‑aware job orchestration.
    • Assumptions/Dependencies: Statutory basis at state/utility level; measurement of contracted vs. realized load; may favor larger incumbents.
  • Bold: Point‑of‑sale consumption tax on AI API services
    • What: Small ad valorem levy on model/API usage (tokens, inference minutes) collected by the provider at billing time.
    • Sectors: Software/SaaS, Finance/FinOps, Policy.
    • Tools/Workflows: Line‑item tax in invoices; provider SDKs for usage accounting; remittance portals akin to sales tax.
    • Assumptions/Dependencies: Clear scoping of “AI service”; harmonization across jurisdictions; managing burden on startups and SMEs.
  • Bold: Creator‑protection tax credits and rate reductions
    • What: Credits or reduced AI tax rates for firms that (i) license datasets, (ii) disclose training data sources, or (iii) pay into creator compensation schemes.
    • Sectors: Creative Industries, Software, Policy.
    • Tools/Workflows: Licensing registries; training‑data disclosure attestations; compensation portals and micropayments.
    • Assumptions/Dependencies: Verification standards to prevent gaming; unsettled fair‑use jurisprudence; admin capacity to audit claims.
  • Bold: Corporate income surtax on AI‑related profits with earmarks
    • What: Surtax on profits attributable to AI products/services to fund worker transition, upskilling, and local infrastructure.
    • Sectors: Finance/Tax, Labor/Education, Policy.
    • Tools/Workflows: Segment reporting for AI profit attribution; grant/voucher distribution systems; program evaluation.
    • Assumptions/Dependencies: Defining “AI‑related profits”; potential incidence on consumers; international profit shifting risks.
  • Bold: Regulatory user fees to fund AI oversight capacity
    • What: Fees on frontier model submissions, safety evaluations, or high‑risk deployments to finance audits, incident reporting, and standards.
    • Sectors: Policy/Regulation, Academia, Safety Labs.
    • Tools/Workflows: Fee schedules; evaluator grant programs; incident databases; independent auditing frameworks.
    • Assumptions/Dependencies: Legal authority; avoiding capture by large firms; maintaining independence of evaluators.
  • Bold: Local community benefit agreements (CBAs) funded by AI taxes
    • What: Statutory or negotiated revenue shares from AI‑linked taxes to support housing, schools, and environmental mitigation near data centers.
    • Sectors: Community Development, Policy, Energy.
    • Tools/Workflows: CBA templates; public dashboards tracking inflows/outflows; participatory budgeting.
    • Assumptions/Dependencies: Transparent governance; enforceable commitments; equitable distribution across affected residents.
  • Bold: Procurement‑linked tax incentives for safe and fair AI
    • What: Tax credits or reduced rates for products passing recognized safety/fairness benchmarks and transparency audits; aligned with public procurement.
    • Sectors: Public Sector, Software, Safety/AI Assurance.
    • Tools/Workflows: Conformity assessments; published model cards/eval reports; registry of certified systems.
    • Assumptions/Dependencies: Robust, scalable audits; clarity on acceptable benchmarks; avoiding perverse incentives.
  • Bold: Public incident/harm monitoring funded by AI tax revenue
    • What: Finance national or state incident reporting portals and monitoring (e.g., chatbot mental‑health incidents, bias reports).
    • Sectors: Healthcare, Public Safety, Policy, Academia.
    • Tools/Workflows: Reporting portals; taxonomy and triage workflows; researcher access with privacy safeguards.
    • Assumptions/Dependencies: Sustained funding; privacy compliance; incentives for reporting and data quality.

Long‑Term Applications

These require further research, measurement standards, or multilateral coordination before wide deployment.

  • Bold: Compute‑based excise (per FLOP/GPU‑hour) with trusted metering
    • What: Tax tied to actual compute used for training/inference to target AI activity directly rather than proxy inputs.
    • Sectors: Cloud, Semiconductors, Software.
    • Tools/Workflows: Hardware attestation, secure telemetry, standardized “AI usage” meters, third‑party verification.
    • Assumptions/Dependencies: Preventing gaming (e.g., throttling, reclassification); privacy of workloads; international standards.
  • Bold: Global coordination on AI tax base and incidence (BEPS‑style)
    • What: OECD‑like framework to reduce leakage and profit shifting for mobile, intangible AI activity.
    • Sectors: International Policy, Finance/Tax.
    • Tools/Workflows: Multilateral agreements; nexus and allocation rules for AI value; dispute resolution mechanisms.
    • Assumptions/Dependencies: Political will; shared definitions of “AI activity” and rent; alignment with digital services tax reforms.
  • Bold: AI sovereign wealth or public equity funds
    • What: Equity, windfall, or rent‑based taxes financing a national fund to share AI gains (e.g., citizen dividends).
    • Sectors: Finance, Policy, Social Welfare.
    • Tools/Workflows: Fund governance, dividend rails, valuation of in‑kind equity, risk management.
    • Assumptions/Dependencies: Constitutional/appropriations constraints; market impacts; rent measurement and timing.
  • Bold: Creator dividend via rent taxes and data unions
    • What: Sectoral rent taxes on generative AI to finance ongoing royalty‑like payments to creators via unions or trusts.
    • Sectors: Creative Economy, Fintech, Software.
    • Tools/Workflows: Data provenance/attribution; collective licensing; micropayment infrastructure and wallets.
    • Assumptions/Dependencies: Reliable provenance tech; opt‑in frameworks; resolution of copyright/fair‑use disputes.
  • Bold: Risk‑tiered excise rates linked to model safety certification
    • What: Lower rates for verified low‑risk models; higher rates for unassessed/high‑risk systems to incentivize evaluations.
    • Sectors: Safety/Assurance, Software, Policy.
    • Tools/Workflows: Standardized risk classes; conformity assessment bodies; periodic re‑certification.
    • Assumptions/Dependencies: Valid, tamper‑resistant risk metrics; avoiding stifling benign innovation.
  • Bold: Privacy externality fees with credits for privacy‑preserving ML
    • What: Fees for training on sensitive personal data without consent; credits for differential privacy, federated learning, or robust de‑identification.
    • Sectors: Health, Finance, Software.
    • Tools/Workflows: Privacy budget meters; DP audits; consent and data‑use registries.
    • Assumptions/Dependencies: Auditable privacy guarantees; harmonized data‑protection rules; avoiding burdens on research use.
  • Bold: Misinformation levy funding content integrity infrastructure
    • What: Levies on high‑scale content generators to finance watermarking, provenance (e.g., C2PA), and detection research.
    • Sectors: Media/Platforms, Advertising, Software.
    • Tools/Workflows: Watermark registries; provenance verification in browsers and platforms; red‑teaming grants.
    • Assumptions/Dependencies: Effective, robust watermarking; platform adoption; risk of speech and rights concerns.
  • Bold: Catastrophic‑risk insurance pool funded by AI hazard levies
    • What: Industry‑wide fund to backstop low‑probability, high‑impact AI failures; pricing encourages safer development.
    • Sectors: Insurance, Policy, Safety.
    • Tools/Workflows: Parametric triggers; risk models; capital reserve governance; stress testing.
    • Assumptions/Dependencies: Actuarial feasibility; correlated‑risk modeling; statutory authority.
  • Bold: Grid‑interactive tax credits for flexible AI workloads
    • What: Credits for shifting training/inference to low‑demand periods or renewable surplus; penalties for peak stress.
    • Sectors: Energy, Cloud/Datacenters, Software.
    • Tools/Workflows: Grid APIs; carbon‑aware schedulers; time‑of‑use telemetry; SLAs reflecting flexibility.
    • Assumptions/Dependencies: Utility coordination; standardized emissions accounting; verifiable scheduling.
  • Bold: International labor adjustment and regional transition funds
    • What: Rent/windfall taxes financing retraining, mobility support, and regional development in areas hit by AI displacement.
    • Sectors: Labor/Education, Economic Development, Policy.
    • Tools/Workflows: Skills mapping; outcome‑based training grants; public‑private placement pipelines.
    • Assumptions/Dependencies: Measuring displacement vs. augmentation; program efficacy; political durability.
View Paper Prompt  View All Prompts 

Glossary

  • Ad valorem: A tax calculated as a percentage of the value of the taxed item. "whether as a percentage of value (an {ad valorem} rate) or a fixed amount per unit (a {per-unit} charge)"
  • Budget reconciliation: A U.S. congressional process that allows budget-related legislation, including tax bills, to pass with a simple majority. "This pathway---known as budget reconciliation---makes federal tax legislation a more viable path than many non-fiscal regulatory proposals when Congress is otherwise gridlocked."
  • Conformity assessments: Evaluations to determine whether systems or processes meet specified standards or regulatory requirements. "conformity assessments"
  • Corrective taxation: Taxes designed to change behavior by pricing socially harmful activities so that private actors internalize their external costs. "including revenue-raising and corrective taxation."
  • Data center tariff: A specialized utility rate structure for data centers that sets terms and prices for electricity usage. "the Public Utilities Commission's ``data center tariff'' reduces risks of speculative data center grid costs from being shifted onto other ratepayers."
  • Data unions: Collective organizations that pool individuals’ data rights to negotiate terms of use and compensation. "many artists and creators have begun exploring ways to protect and preserve their work via data unions"
  • Demand externality: A spillover effect where reduced income (e.g., from layoffs) lowers aggregate demand, imposing costs on others not accounted for by the firm. "argues that AI layoffs create a demand externality, motivating a Pigouvian automation tax because firms do not internalize the broader loss of consumer demand from displaced workers."
  • Earmarking: Dedicating tax revenues to a specific purpose rather than the general budget. "If desired, tax revenue should be earmarked."
  • Excess-profit tax: A levy on profits above a normal or benchmark return, often used during extraordinary windfalls. "Taxpayer and instrument: Who owes the tax and through which mechanism (excise, consumption, corporate, payroll, excess-profit, windfall-profit, rent)."
  • Excise tax: A tax imposed on a specific good, service, or activity. "including excise taxes, payroll taxes, consumption taxes, and corporate or rent-based taxes."
  • Externality: An uncompensated cost (or benefit) imposed on third parties that is not fully reflected in market prices. "Externalities are uncompensated costs borne by third parties and not fully reflected in market prices."
  • Fair use: A copyright doctrine permitting certain uses of protected works without permission under defined circumstances. "the court held that their use for model training constituted fair use"
  • Hyperscale data centers: Very large computing facilities designed to scale massively in capacity, typically supporting AI training and inference at scale. "The fast expansion of hyperscale data centers has brought with it higher electricity and water charges for residents living near them"
  • Jurisdictional leakage: The shifting of economic activity to other regions to avoid taxes or regulations, undermining policy effectiveness. "incidence effects, jurisdictional leakage, political capture, and potential costs to innovation and competitiveness."
  • Marginal social cost: The additional total cost to society of producing one more unit of an activity, including externalities. "Pigouvian taxes aim to align supply with marginal social cost."
  • Per-unit charge: A fixed tax amount applied to each unit of the taxed base. "a {per-unit} charge"
  • Pigouvian correction: Using taxes or subsidies to correct market failures by aligning private incentives with social costs or benefits. "AI taxation should not be understood only as Pigouvian correction."
  • Regulatory capacity: The resources, expertise, and infrastructure required by government to monitor, enforce, and implement regulation. "The third is to fund regulatory capacity, especially AI oversight."
  • Regulatory sandbox: A controlled environment that allows limited testing of innovations under oversight and relaxed rules. "regulatory sandboxes"
  • Remitter: The party responsible for collecting and transferring a tax to the government. "while the tax remitter, is the party legally required to transfer payment to the government"
  • Rent-based tax: A tax on economic rents—returns above the normal required rate—often designed to minimize distortions. "including corporate income and rent-based taxes, consumption taxes on AI-related services, and excise instruments tied to specific AI activities."
  • Sovereign wealth fund: A state-owned investment fund that manages public assets for long-term objectives. "a public sovereign wealth fund"
  • Tax base: The specific item, activity, or amount on which a tax is levied. "The tax base is the economic object to which a tax applies"
  • Tax incidence: The distribution of the economic burden of a tax, irrespective of who is legally liable. "Finally, tax incidence describes who ultimately bears the economic burden of a tax after prices, wages, and contracts adjust"
  • Tax instrument: The legal form or mechanism through which a tax is imposed. "The tax instrument refers the legal form of the levy---such as an income, excise, or payroll tax."
  • Tax neutrality: The principle that the tax system should not unduly favor certain economic choices over others. "restoring tax neutrality between workers and machines"
  • Taxpayer: The entity legally liable for the tax under the law. "The taxpayer, by contrast, is the party on whom the law imposes tax liability"
  • User fee (regulatory user fee): A charge to cover the costs of regulatory oversight or services provided by the government. "Excise tax, consumption tax, corporate income tax, rent tax, payroll tax, or regulatory user fee."
  • Windfall-profit tax: A tax on unexpectedly large or extraordinary profits, often tied to unusual market conditions. "Taxpayer and instrument: Who owes the tax and through which mechanism (excise, consumption, corporate, payroll, excess-profit, windfall-profit, rent)."
View Paper Prompt  View All Prompts 

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Generate Now

Collections

Sign up for free to add this paper to one or more collections.

Sign Up

Tweets

Sign up for free to view the 1 tweet with 0 likes about this paper.

Sign Up for Free

HackerNews

  1. Taxing Artificial Intelligence (11 points, 0 comments) 
  2. Taxing Artificial Intelligence (4 points, 1 comment)