Labor market impacts of AI: A new measure and early evidence
Anthropic introduces observed exposure, a labor-market measure that combines theoretical LLM capability, Claude usage data, work context, automation weight, and task-time shares to track early AI exposure by occupation.
Core Thesis
The report argues that AI labor-market effects are not yet visible as broad unemployment shocks, so better leading measures are needed. Observed exposure links O*NET occupational tasks, Anthropic Economic Index usage, and prior task-level LLM capability estimates. It finds that actual AI coverage remains far below theoretical capability, that more exposed occupations have weaker BLS projected growth, and that highly exposed workers tend to be older, female, more educated, and higher paid. The unemployment evidence is currently small and statistically weak, but hiring into exposed occupations for workers aged 22-25 shows a tentative slowdown.
Argument Structure
The infographic follows the structure of the generated knowledge graph: section claims, glossary entities, a how-to interpretation path, and linked FAQ nodes.
Observed exposure as a new measure
The report defines observed exposure as a task-to-occupation coverage measure that weights real work usage, automation, and task importance.
Observed exposure, Theoretical LLM capability, Anthropic Economic Index
Data sources and task coverage
The measure combines O*NET task data, Anthropic Economic Index usage, and Eloundou et al. task-level LLM exposure estimates.
Gap between capability and adoption
Actual professional Claude use remains a fraction of theoretical capability, with Computer and Math at 33 percent observed coverage versus much higher theoretical scope.
Coverage gap, Computer and Math category, Automated use patterns
Most exposed occupations
Computer Programmers, Customer Service Representatives, and Data Entry Keyers rank among the most exposed occupations under the observed coverage measure.
Computer Programmers, Customer Service Representatives, Data Entry Keyers
BLS projections and worker characteristics
Higher observed exposure is associated with slightly weaker BLS projected growth and with a worker profile that is older, female, more educated, and higher paid.
BLS employment projections, Current Population Survey, High-exposure workers
Initial unemployment and hiring evidence
The report finds no systematic unemployment increase for exposed workers, but sees suggestive evidence of reduced hiring into exposed occupations for young workers.
Difference-in-differences, Young-worker hiring, Job finding rate
Why the framework matters
Anthropic frames the work as a repeatable measurement system that can be updated as AI usage, capability, and employment data evolve.
How The Argument Progresses
The knowledge graph models the article as an explicit sequence of reasoning steps rather than a loose summary.
1
Start with task-level capability
Use prior estimates of which occupational tasks LLMs can theoretically speed up.
2
Filter through observed work usage
Use Anthropic Economic Index data to identify which tasks appear in Claude usage in professional settings.
3
Weight automation and task importance
Give automated uses more weight than augmentation and aggregate task coverage by time spent in each occupation.
4
Compare exposure with projections
Test whether observed exposure aligns with BLS projected employment growth and worker characteristics.
5
Track unemployment and hiring
Use CPS data to study unemployment gaps and young-worker job starts in exposed versus unexposed occupations.
Glossary From The Graph
These linked entities are exposed as DefinedTerm nodes in the RDF and mirrored in the embedded JSON-LD.
Observed exposure
Anthropic's new measure of occupational AI exposure based on theoretical capability, observed usage, automation weight, work context, and task-time shares.
Theoretical LLM capability
A task-level estimate of whether an LLM could make a task substantially faster, before accounting for real-world adoption.
Anthropic Economic Index
Anthropic's usage-based research program for studying which economic tasks are performed with Claude.
O*NET database
The US occupational database that enumerates tasks associated with roughly 800 occupations.
Task-level exposure
A task-level estimate of how much AI could affect work before aggregation to occupations.
Claude usage data
Observed Claude traffic used to determine which theoretically exposed tasks are actually appearing in work-related settings.
Automated use patterns
Usage patterns that receive higher weight than augmentative use because they more directly indicate task substitution potential.
Coverage gap
The gap between tasks that LLMs could theoretically help with and tasks where professional AI usage is actually observed.
FAQ From The Knowledge Graph
Each question and answer below is linked to a separate resolver-backed node and mirrored in the metadata graph.
What does the report say about current AI coverage?
It says actual coverage remains only a fraction of what is theoretically feasible.
Who is more likely to be in exposed occupations?
The report says exposed workers are more likely to be older, female, more educated, and higher paid.
Does the report find higher unemployment from AI?
It finds no systematic unemployment increase for highly exposed workers since late 2022.
What early labor-market signal does the report find?
It finds suggestive evidence that hiring of workers aged 22-25 into exposed occupations has slowed.