Charging levies on AI 'tokens'—a usage tax applied at the point of inference—could give governments a practical way to tax AI-generated value and blunt automation-driven revenue and wage shocks; the approach hinges on robust cryptographic receipts, norm-based rates, white-box audits, and difficult international cooperation.

Main Finding

The paper argues that “token taxes” — a usage-based surcharge on model inference tokens charged at the point of sale — are a practical, enforceable policy instrument to mitigate major economic risks from AGI (lost tax bases, citizen disempowerment, and widening global inequality). Token taxes exploit existing compute-governance infrastructure, capture value where AI is used (not only where models are hosted), and can be audited via a staged pipeline to limit evasion.

Key Points

• Definition and mechanism

  • Token tax = percentage surcharge on billed model tokens (e.g., a 10% markup on cost-per-token) remitted to government.
  • Collected at point of sale; compute providers act as intermediaries to record/verify and remit tax.

• Two claimed advantages over prior “robot taxes”

  • Enforceability: build on current cloud/compute telemetry and a proposed 3-stage audit pipeline to detect or deter misreporting.
  • Equity/allocative: taxes are levied where tokens are consumed, helping ensure AI-consuming jurisdictions (including lower-income “Compute South” countries) receive revenue rather than revenue accruing only to hosting jurisdictions (“Compute North”).

• Staged audit pipeline (enforcement design)

• Black-box token audits: cloud providers log token-level usage and verify reported token counts against independent logs.
• Norm-based taxes: if reporting is suspect, apply empirically derived norm rates per model category (a default flat tax based on typical token usage).

• White-box audits: require disclosure of internal generative-process logs to third-party auditors (legal requirement when other stages fail).

• Alternatives and complements

  • FLOP (floating-point ops) taxes: tax on compute is an alternative and is already used as a regulatory proxy in some legislation. Authors propose token and FLOP taxes can be complementary (hybrid).
  • Concerns addressed: possible innovation disincentives and firm relocation; geopolitical veto by AI superpowers (mitigation via coalitions of the willing); distributive effects.

• Research & policy roadmap

  • Recommend agent-based modeling (ABM), including LLM-based ABMs, to forecast token-tax impacts under different AGI growth scenarios.
  • Need for standards and legal mandates obliging compute providers to collect and report token telemetry.

• Key objections acknowledged

  • Disincentive to innovation and capital flight to low-tax jurisdictions.
  • Measurement/evasion risk (providers or model hosts under-reporting tokens).
  • Geopolitical refusal by major AI-supply states; mitigation via regional/coalition coordination.

Data & Methods

• Nature of the paper

  • Conceptual/policy design paper; no original empirical dataset or statistical estimation is presented.
  • Synthesis of prior empirical findings and governance literature (examples cited in paper include studies on AI exposure and unemployment, compute concentration, cloud telemetry capabilities, and auditing techniques).

• Empirical evidence drawn on

  • Early labor-market evidence (e.g., elevated unemployment in AI-exposed roles).
  • Compute concentration evidence (Compute North vs. Compute South).
  • Prior work documenting the possibility of token under-counting and approaches to infer hidden reasoning tokens.

• Proposed technical/policy methods (for implementation and evaluation)

  • Enforcement architecture: use cloud hyperscaler telemetry as the authoritative record for token usage; mandate token-level logging and reporting.
  • Audit methodology: implement the three-stage pipeline (black-box logs → norm-based defaults → white-box disclosure).
  • Evaluation method: agent-based models (ABMs) to simulate market responses, firm relocation incentives, distributional outcomes, and dynamic fiscal effects under alternative token-tax designs. The authors recommend collaboration between governance researchers, economists, and technical teams to build realistic ABMs that capture firm behavior, cross-border choices, and policy interactions.

• Limitations of current methods identified by authors

  • Black-box logs can be manipulated unless legal/contractual obligations bind providers.
  • Norm-based rates require robust, continuously updated empirical baselines per model category.
  • White-box audits raise proprietary/competition concerns and require legal/regulatory authority.

Implications for AI Economics

• Fiscal resilience and tax neutrality

  • Token taxes give governments a direct lever to tax AI-generated economic value, helping to replace eroding labor tax bases and restore a form of tax neutrality as capital (models) performs labor tasks.

• Distributional and international equity

  • Because token taxes are collected where consumption occurs, they can redistribute revenue toward AI-consuming countries and mitigate some global inequality risks from compute concentration.

• Market incentives and innovation trade-offs

  • Token taxes alter the marginal cost of delivering AI services. Potential consequences:
    • Firms may pass costs to consumers, reduce model use, or change product design to lower taxed token usage (potentially desirable if it curbs harmful overuse, but could also reduce beneficial services).
    • Risk of tax-driven geographic relocation of firms or compute infrastructure; requires international coordination to avoid a race-to-the-bottom.

• Enforcement and governance feasibility

  • Leveraging cloud provider telemetry is a pragmatic enforcement path, but it depends on legal mandates, standardization of token accounting, and the willingness of hyperscalers to cooperate.
  • Norm-based fallback and white-box audits provide layered deterrence but raise operational, privacy, and commercial-friction questions.

• Policy design & research priorities

  • Need for careful calibration: tax rates should be tested (via ABMs and empirical pilots) to balance revenue goals against innovation and distributional effects.
  • Hybrid policy mixes (token + FLOP taxes, corporate taxes, direct transfers/UBI funded by token taxes) should be modeled to assess welfare implications and optimal design.
  • International coordination (coalitions of the willing, trade/regulatory agreements) is crucial to prevent vetoes or circumvention by AI superpowers.

• Practical next steps recommended

  • Develop technical standards for token accounting and independent logging.
  • Build ABMs incorporating firm-level decisions, cross-border compute choices, consumer demand sensitivity to token prices, and public finance feedbacks.
  • Pilot token-tax reporting regimes with voluntary participants or regional coalitions to gather empirical data before wide rollout.

Overall, the paper frames token taxes as a technically feasible, enforceable policy instrument worth prioritizing in AGI-era fiscal governance, while highlighting important open research questions on economic impacts, enforcement design, and international coordination.