U.S. Community and social service Occupations · Share of Work Robots Can Do

## What this measures

The share of working time in U.S. community and social service occupations (SOC 21) spent on tasks that a robot can already perform in at least a purpose-built setting, in percent. From Anthropic's "What work can robots do?" (September 30, 2026): each O*NET task was rated for the least controlled environment in which a cited, existing robot could do it, from E0 (no robot can) to E3 (an unstructured setting such as a home or a construction site). A task counts as exposed at E1 or above.

## Why it matters

It shows which kinds of work today's robots could already take on, which is where automation pressure on physical jobs will come first if robot costs keep falling.

## How to read it

Percent of working time. Within an occupation, tasks are weighted by the share of working time they take (estimated by Claude in Anthropic's release). Occupations are combined into SOC major groups, weighted by U.S. employment (BLS Occupational Employment and Wage Statistics, May 2024). Across all U.S. work the figure is about a third, the share the report gives ("74% of physical work, or 34% of all work").

## What it does not say

Exposure is technical feasibility, not adoption: most exposed tasks need a purpose-built or tightly controlled setting, and the report finds robots cost-competitive for only a small share of work. Non-physical tasks count as unexposed here; the report's separate measure of exposure to language models is not included. One release, dated its publication day; it does not update.

## History

Single release: one value per occupation group from one Anthropic release, so there is no earlier series to backfill. A new point is added when Anthropic publishes a new release.

## Source

Anthropic Economic Index, robot_exposure release (huggingface.co/datasets/Anthropic/EconomicIndex), data released under CC BY 4.0. Includes information from the O*NET 29.3 Database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA), used under CC BY 4.0; O*NET® is a trademark of USDOL/ETA; Anthropic has modified this information and USDOL/ETA has not approved, endorsed, or tested these modifications. Employment weights: BLS OEWS (public domain). Aggregation by Palanor.