Anthropic Economic Policy Responses for AI

Anthropic has proposed a framework of nine policy ideas designed to help governments manage the economic transitions caused by advanced AI. Because the scale and speed of AI adoption are uncertain, the company suggests matching specific policy responses to three distinct economic scenarios: modest disruption, moderate acceleration, and fast-moving transformation.

Policies for Nearly All Scenarios

These proposals are intended for use regardless of whether AI's negative effects on the labor market are modest or significant. They focus on foundational workforce development and infrastructure efficiency.

Workforce Training Grants

To support on-the-job training, the "Workforce Training Grant" proposal suggests providing substantial annual subsidies (e.g., $10,000 per year in the US) directly to employers who establish structured training programs. Funding could be sourced by redirecting existing higher education subsidies or implementing taxes on AI consumption.

Tax Incentives for Retention and Retraining

Current tax codes may bias physical capital investment over human capital. Proposed reforms include eliminating the $5,250 cap on tax-free educational assistance and extending full, immediate expensing to all job-related training to make retraining more financially attractive than layoffs.

Closing Corporate Tax Loopholes

To prevent AI transformation from straining government budgets, proposals include closing the "partnership gap" to prevent entity-level tax avoidance and implementing market-based apportionment. The latter would allocate business taxes based on customer locations to better capture value from digital and intangibles-based business models.

AI Infrastructure Permitting Reform

Accelerating the construction of data centers, transmission infrastructure, and power generation is critical for productivity and national security. Proposed reforms include updating the National Environmental Policy Act (NEPA) to allow for advance analyses of certain facilities and leveraging federal authorities to expedite critical transmission buildouts.

Policies for Moderate Acceleration Scenarios

These policies are designed for scenarios where AI leads to measurable wage declines and job losses for large portions of the workforce.

AI Displacement Assistance

Adapting the Trade Adjustment Assistance (TAA) model, an "Automation Adjustment Assistance" (AAA) program would provide skills training and support for workers displaced by AI. Initial funding could be set at approximately $700 million annually, potentially scaling upward via taxes on AI-driven revenues from firms with high market capitalization.

Compute and Token Taxes

Economists suggest studying taxes on token generation, robots, and digital services. A "token tax" on AI-generated tokens sold to end users may be effective while humans remain the dominant consumers. If AI systems become the primary consumers of resources, taxes on compute and hardware accumulation may be the only way to capture AI-generated windfalls.

Policies for Fast-Moving Scenarios

These proposals address scenarios involving dramatic job losses and worsening inequality.

National Sovereign Wealth Funds

To distribute AI-derived wealth more equitably, governments could create sovereign wealth funds to acquire stakes in AI-related assets. A similar concept, the "AI Bond" proposed for the UK, would aim to ensure investment in the AI stack and distribute returns broadly across the population.

Modernized Consumption Taxes

As the share of labor in economic production declines, income taxes may become insufficient. Proposals include adopting or modernizing Value-Added Taxes (VAT) to shift the tax burden toward consumption. Additionally, a "low-rate business wealth tax" could complement income taxes to reduce accounting manipulation and capture value from highly profitable enterprises.

Research and Implementation

Anthropic has committed $10 million to its Economic Futures Program to support empirical research on these policy ideas and host symposia for policymakers and researchers. The company emphasizes that these proposals are starting points for research and debate rather than definitive policy recommendations.