Anthropic Introduces ‘Observed Exposure’ Framework

Anthropric has published a new measure of AI displacement risk called observed exposure, which combines theoretical LLM capability with real-world usage data. The framework weights automated and work-related uses more heavily than augmentative uses, offering a more granular picture of occupational vulnerability than capability estimates alone.

High Alignment Between Theory and Practice

97% of tasks observed across the previous four Anthropic Economic Index reports fall into categories rated as theoretically feasible by Eloundou et al. (β=0.5 or β=1.0). More strikingly, tasks rated β=1 (fully feasible for an LLM alone) account for 68% of observed Claude usage, while tasks rated β=0 (not feasible) account for just 3% of observed Claude usage.

Occupations at Highest Risk

Computer Programmers emerge as the most AI-exposed occupation by Anthropic’s measure, with 75% task coverage. Customer Service Representatives rank second, with their main tasks increasingly appearing in first-party API traffic. Data Entry Keyers claim third position among the most exposed, with 67% task coverage driven primarily by automation of reading source documents and entering data.

Theoretical LLM feasibility spans 90% of tasks in Office & Admin occupations and 94% in Computer & Math roles—yet Claude currently covers just 33% of all tasks in the Computer & Math category, suggesting substantial room for capability expansion.

Workers with Zero AI Exposure

Approximately 30% of workers have zero AI task coverage under Anthropic’s measure, including Cooks, Motorcycle Mechanics, Lifeguards, Bartenders, Dishwashers, and Dressing Room Attendants.

Employment Growth Correlation

For every 10 percentage point increase in observed AI exposure, the BLS employment growth projection for that occupation drops by 0.6 percentage points. The BLS 2025 employment projections cover predicted changes in employment for every occupation from 2024 to 2034.

Demographic Disparities in AI Exposure

Workers in the top quartile of AI exposure face markedly different demographic profiles than those with zero exposure. Workers in the top quartile are 16 percentage points more likely to be female and 11 percentage points more likely to be white. They earn 47% more on average.

Education gaps are even starker: workers with graduate degrees represent 4.5% of the zero-exposure group but 17.4% of the most AI-exposed group—an almost fourfold difference.

Early Signs of Labor Market Slowdown

Anthropric found no systematic increase in unemployment for workers in the most AI-exposed occupations since late 2022, but early hiring signals are concerning. The research identified a 14% drop in the job-finding rate for workers aged 22–25 entering AI-exposed occupations in the post-ChatGPT era compared to 2022, though this result is described as just barely statistically significant.

Job finding rates for low/zero AI-exposure occupations remain stable at approximately 2% per month for workers aged 22–25, while entry into the most AI-exposed jobs has decreased by about half a percentage point. Critically, the decline in job-finding rates for AI-exposed occupations is observed only for workers aged 22–25; there is no such decrease for workers older than 25.

Brynjolfsson et al. (2025) corroborate this concern, reporting a 6–16% fall in employment in AI-exposed occupations among workers aged 22 to 25, attributed primarily to a slowdown in hiring rather than an increase in separations.

Real-World Impact: Call Centre Employment

Goldman Sachs analysis provides concrete evidence of AI’s labor market effect in specific sectors. Call centre employment in the US is now 39% below pre-AI trend levels, while Canada has seen a 33% decline and Germany a 27% decline.

Goldman Sachs combined 11 surveys measuring AI adoption across countries and found that major developed markets have AI adoption rates of roughly 15% to 20%. France, the US, the Netherlands, and the UK are leading AI adoption among developed economies, while Italy, Japan, and New Zealand are among those at the lower end. Major emerging markets have estimated AI adoption rates of between 10% and 15%.

Historical Context

A prior attempt to measure job offshorability identified roughly a quarter of US jobs as vulnerable, but a decade later most of those jobs maintained healthy employment growth. Whether AI-exposed roles follow a similar trajectory remains an open question as early data emerges.

Source: Anthropic