Anthropic unveils AI job exposure index

Anthropic, the maker of the Claude AI model, has launched an index that it says is an early-warning system for potential AI-driven white-collar job losses.

The index scores occupations by task-level exposure using estimates of which tasks large language models can perform and by measuring which tasks are already automated in Anthropic’s anonymized data.

Anthropic says computer programmers show about 75% task coverage and that roughly 30% of occupations do not meet its threshold for being exposed.

So far Anthropic finds limited evidence that exposed occupations have faced higher unemployment, but reports suggestive signs that hiring of workers ages 22 to 25 has slowed in exposed roles.

The company says the measure is meant to help economists detect ambiguous or delayed labour effects that aggregate statistics might miss and it notes government surveys are also changing how they record AI use.

🔗 Source: Axios

🧠 Food for thought

Implications, context, and why it matters.

Real-world AI use lags far behind its theoretical potential

Anthropic reports that programmers reach about 75% task coverage in its index. The index counts real Claude AI model use from anonymized data, so it can sit far below what AI could do in theory.
In computer and math jobs, 94% of tasks are theoretically exposed to AI. Claude currently appears in 33% of tasks in that group 1.
Projected job growth does not move with theoretical exposure. Observed AI use has a modest negative correlation with projected job growth 1.
Workers in the most exposed roles are higher paid. They earn 47% more than unexposed workers, plus they are nearly four times likelier to hold a graduate degree 1.

AI’s economic impact is shifting from job replacement to job transformation

Anthropic frames its index as a heads-up tool for potential AI-driven white-collar job losses, and it tracks unemployment. U.S. Census Bureau surveys report that adopting technology, including AI, has had little impact on total worker numbers 2.
Workplace AI use may be leveling off. One survey records worker usage dropping from 46% to 37% between June and September 2025 3.
Near-term effects look like task reshuffling inside jobs. AI use clusters in tasks tied to more education, averaging 14.4 years of schooling versus the 13.2-year economy-wide average 4.
Technical writers offer one example. AI can take on complex analysis, leaving more routine work for writers and changing how expertise is valued 4.