An AI company funding independent economic research, hosting policy symposia, and building public data infrastructure sounds like an unusual strategy. It is. But Anthropic's Economic Futures program represents a specific bet: that understanding AI's economic impact requires more than internal analysis. It requires an ecosystem of measurement, and that ecosystem does not exist yet. Anthropic is building it.
Why an AI Lab Is Investing in Economics
The premise is straightforward. AI systems are already reshaping how people work, yet the research community lacks the data to understand the pace, distribution, and character of that transformation. Academic studies of technology's labor market impact typically use government survey data that lags by months or years. Industry reports rely on self-reported adoption surveys with unknown biases. Neither captures what is happening on the ground right now.
Anthropic has something most researchers do not: real-time data on how millions of people use AI in professional contexts. Through Clio, a privacy-preserving analytics system, the company can observe patterns in Claude conversations at scale. This data maps directly to occupational tasks, industries, and use cases.
But having data is not the same as having understanding. And Anthropic recognizes that its own analysis, however rigorous, carries inherent limitations. Claude users are not representative of the global workforce. The platform captures AI usage, not AI impact. Observed behavior reflects adoption, not displacement.
This is where the Economic Futures program enters. Launched in late 2025 and expanded through 2026, it positions Anthropic not as the authority on AI's economic effects, but as a platform for building the knowledge infrastructure the field needs.
The Three Pillars
The program operates through three interconnected pillars, each addressing a different gap in the current research landscape.
Pillar 1: Catalyzing Independent Research
Through grants ranging from $10,000 to $50,000, API credits, and strategic partnerships, the program funds external researchers studying AI's economic effects. The research questions they prioritize reveal the program's intellectual agenda:
Labor market transitions. How quickly are AI-enabled jobs emerging relative to job displacement? What are the actual flows of workers between occupations as AI adoption increases? Which sectors are seeing net job creation versus reduction?
Human-AI complementarity. Which human skills will remain valuable as AI advances? How should education and professional licensing evolve? What new capabilities emerge when humans work alongside AI systems?
Productivity measurement. How can we capture AI-driven output changes that traditional GDP metrics miss? What is the relationship between AI adoption and measured productivity at the firm and economy level?
Market structure and competition. How does AI adoption affect market concentration, firm dynamics, and competitive advantage? Does AI disproportionately benefit large firms or small ones?
The decision to fund external research rather than keeping all analysis in-house is deliberate. It addresses the credibility problem head-on: an AI company's own assessment of its economic impact carries an inherent conflict of interest. Independent researchers using Anthropic's data can reach conclusions that Anthropic's own team might not publish.
Pillar 2: Collaborative Policy Development
The Economic Futures Symposia bring together researchers, policymakers, and practitioners to evaluate policy proposals backed by real-world data. The first symposium in Washington, D.C. produced a collection of policy proposals that reveal the range of thinking:
Abigail Ball from American Compass proposed workforce training grants targeted at AI-displaced workers. Andreas Kern from Georgetown explored how AI could serve emerging market monetary authorities. Anh Nguyen and Era Dabla-Norris from the IMF outlined fiscal policies for managing the AI transition. David Gamage from the University of Missouri proposed tax architecture for a resilient AI economy. Neil Thompson from MIT introduced an "expertise framework" for calibrating labor market policy responses.
The diversity of these proposals illustrates a core challenge: there is no consensus on what "good" AI economic policy looks like. The symposia do not aim to produce consensus. They aim to ensure that policy debates are grounded in data rather than speculation.
A European symposium followed, hosted at the London School of Economics, expanding the geographic and disciplinary reach.
Pillar 3: Expanding Measurement and Data Infrastructure
The Anthropic Economic Index is the program's most concrete output. It is a recurring data report analyzing how people use Claude across the economy, mapping conversations to professional tasks and occupational categories.
The Index has evolved significantly since its February 2025 debut. The January 2026 report introduced the "economic primitives" framework: five foundational metrics (task complexity, skill level, use case, AI autonomy, and task success) that standardize measurement across reports. The March 2026 report added learning curve analysis, showing how user experience affects task success rates.
The data is publicly available. Anthropic publishes datasets on Hugging Face, enabling any researcher to analyze the underlying data independently. The Economic Index Survey, launched in April 2026 as a monthly panel survey conducted through Anthropic Interviewer, adds a qualitative dimension: direct accounts from workers about how AI is changing their daily work.
From Data to Framework: The Observed Exposure Metric
The program's most significant methodological contribution is the "observed exposure" metric, developed in the labor market impacts paper by Massenkoff and McCrory (March 2026).
Traditional approaches measure AI's labor market potential by estimating what tasks AI could theoretically perform. Anthropic's approach adds a second layer: what tasks are people actually using AI for. The observed exposure metric combines:
- Theoretical capability from Eloundou et al. (2023), measuring which tasks LLMs can accelerate by at least 2x
- Claude usage data from the Economic Index, measuring which tasks show significant work-related usage
- Automation weighting, giving full weight to fully automated workflows and half weight to augmentative use
- Occupational aggregation, averaging task-level coverage weighted by time spent
The results paint a nuanced picture. Computer programmers show 75% observed exposure. Customer service representatives and data entry keyers also show high rates. But 30% of US workers show zero observed exposure: cooks, motorcycle mechanics, lifeguards, bartenders.
The gap between theoretical capability and observed exposure is consistently 50-65 percentage points across occupational categories. This gap is the program's central object of study. Understanding why it exists and how fast it is closing is essential for calibrating policy responses.
The 81,000-Person Study: Connecting Data to Experience
The qualitative counterpart to the Economic Index's quantitative data came from Anthropic's 81,000-person user study, conducted across 159 countries in 70 languages. This is the largest multilingual qualitative AI user study to date.
The economic analysis of the survey, published in April 2026, produced several findings that complement the Index data:
One-fifth of respondents expressed concern about economic displacement from AI. This concern was not evenly distributed. It correlated strongly with observed exposure: workers in occupations where Claude handles a larger share of tasks were more worried about replacement. Every 10 percentage point increase in observed exposure corresponded to a 1.3 percentage point increase in perceived job threat.
Early-career workers expressed more anxiety than senior employees. This aligns with the labor market paper's finding that hiring rates for 22-25 year olds in high-exposure occupations dropped approximately 14%, a borderline significant result that merits continued monitoring.
The most striking finding was the productivity-anxiety paradox. Respondents who reported the largest speed gains from AI were also the most nervous about AI's effect on their jobs. Speedup and threat perception increased together. Workers who saw their tasks getting faster also saw their roles becoming more replaceable.
The Exponential View Interview: Context from the Economics Lead
Peter McCrory, head of Anthropic's economics team, discussed the program's approach in an interview with Exponential View. Several points from that conversation illuminate the program's philosophy:
"We understand the coal industry better than the AI economy right now." This framing captures the measurement gap the program aims to close. Entire fields of economics have been built around understanding how previous technological transformations affected labor markets, from the steam engine to the internet. AI's economic transformation is happening faster than the research infrastructure can track it.
"Implementation resembles a staircase, not a curve." Adoption does not follow a smooth exponential. It proceeds in bursts as specific task clusters reach threshold levels of reliability and trust. The Economic Index data supports this: usage concentrates heavily in a few tasks, then spreads rapidly once adjacent tasks cross a tipping point.
"The burden of knowledge." McCrory noted that people who understand AI best are often the most uncertain about its economic effects. This is consistent with the survey data showing that technical workers, who have the most direct experience with AI's capabilities, are also the most concerned about displacement.
What Makes This Program Different
Several features distinguish Anthropic's Economic Futures program from other AI industry economic research efforts.
Data transparency. Anthropic publishes its Economic Index data, including anonymized datasets on Hugging Face. Other AI companies collect similar usage data but do not make it publicly available. This transparency enables independent verification and analysis.
Privacy architecture. The entire measurement system is built on Clio, which analyzes conversation patterns without exposing individual conversations. This privacy-preserving approach allows Anthropic to analyze sensitive economic data without compromising user trust.
Multi-disciplinary design. The program integrates quantitative data analysis (Economic Index), qualitative research (81K study, monthly survey), academic funding (research grants), and policy engagement (symposia). This breadth addresses the inherent limitations of any single approach.
Institutional humility. The program explicitly frames Anthropic's data as a starting point, not a definitive answer. The Forbes critique by Hamilton Mann argued that observed exposure is not a general measure of AI's labor market impact, and Anthropic's team has acknowledged this limitation publicly. Funding independent researchers who may challenge Anthropic's own interpretations is an unusual institutional choice.
What to Watch
The program's trajectory suggests several developments worth tracking:
The monthly Economic Index Survey will build a longitudinal dataset tracking how AI's economic impact evolves over time. This time-series data is far more valuable for policy than any single cross-sectional snapshot.
The expansion to UK and European markets, announced in late 2025, broadens the geographic scope beyond the US-centric initial data. Different regulatory environments, labor market structures, and cultural attitudes toward AI will produce different adoption patterns.
The interaction between observed exposure data and traditional labor market statistics. As the Bureau of Labor Statistics and equivalent agencies incorporate AI-specific measurements, the gap between real-time usage data and official statistics should narrow. The program's public datasets make this integration easier for external researchers.
The policy proposals generated through the symposia will increasingly face real-world tests. As EU AI Act compliance requirements take effect in August 2026, the need for evidence-based AI governance frameworks will become more urgent.
FAQ
What is the Anthropic Economic Futures Program? A three-pillar research initiative: funding independent research on AI's economic effects, hosting policy symposia for evidence-based discussion, and expanding the Economic Index measurement infrastructure.
How is this different from the Economic Index? The Economic Index is the data infrastructure (Pillar 3). The Economic Futures Program encompasses the Index plus independent research funding (Pillar 1) and policy symposia (Pillar 2).
Why is an AI company funding this research? Anthropic positions the program as addressing a market failure: the research community lacks real-time data on AI's economic effects. By making its usage data available and funding independent analysis, the company aims to accelerate understanding without being the sole authority.
What is the observed exposure metric? A measure combining theoretical AI task capability with actual Claude usage data to estimate what share of each occupation's tasks are currently being AI-assisted. Computer programmers show 75% observed exposure; 30% of US workers show zero.
How does the 81,000-person study connect to the program? The survey provides qualitative data complementing the Economic Index's quantitative measurements. It shows that worker perceptions of AI risk correlate with measured exposure levels, validating the program's measurement approach.
References
- Anthropic Economic Futures Program: https://www.anthropic.com/economic-futures/program
- Anthropic Economic Index: https://www.anthropic.com/economic-index
- Economic Futures Symposium Proposals: https://www.anthropic.com/economic-futures/symposium-proposals
- Labor Market Impacts Paper: https://www.anthropic.com/research/labor-market-impacts
- What 81,000 People Told Us About the Economics of AI: https://www.anthropic.com/research/81k-economics
- What 81,000 People Want from AI: https://www.anthropic.com/81k-interviews
- Economic Index Primitives: https://www.anthropic.com/research/economic-index-primitives
- Economic Index Survey Announcement: https://www.anthropic.com/research/economic-index-survey-announcement
- Estimating AI Productivity Gains: https://www.anthropic.com/research/estimating-productivity-gains
- Clio Privacy-Preserving Analysis: https://www.anthropic.com/research/clio
- Exponential View Interview with Peter McCrory: https://www.exponentialview.co/p/anthropics-head-of-economics-on-ai
- Economic Index Dataset on Hugging Face: https://huggingface.co/datasets/Anthropic/EconomicIndex