Is your job on the AI firing line? Anthropic lists roles most under threat

Artificial intelligence has not yet led to a measurable rise in unemployment in the United States, but hiring has slowed in some highly exposed professions, especially among young workers. This finding comes from new research that introduces a measure called “observed exposure” to assess how AI may affect jobs and tracks early signals in the labour market.

The study combines theoretical AI capability with real-world usage data and compares those findings with employment trends. While many occupations show some level of exposure, the researchers find limited evidence so far that AI has changed overall employment levels.

New Measure to Track AI Risk

The report introduces “observed exposure” as a new way to measure AI displacement risk. It blends the theoretical capability of large language models (LLMs) with actual usage data from professional settings. The method gives greater weight to automated uses and work-related applications rather than assistive or personal uses.

(Join our ETNRI WhatsApp channel for all the latest updates)


The researchers draw on three main data sources: the O*NET database covering around 800 US occupations, usage data from the Anthropic Economic Index, and earlier task-level exposure estimates that assess whether an LLM can complete a task at least twice as fast. Tasks are scored based on whether they are fully feasible, partially feasible with additional tools, or not feasible.

The study notes that AI remains far from its theoretical potential. Many tasks that could be automated in theory are not yet widely used in practice due to legal limits, software requirements and human oversight. Still, 97% of tasks observed in usage data fall into categories rated as theoretically feasible.

Which Jobs Are Most Exposed

Exposure increases when tasks are theoretically possible with AI, frequently used in professional contexts, automated rather than assistive, and account for a larger share of a worker’s time. The researchers calculate task-level coverage and then aggregate it to the occupation level.

Computer programmers rank highest in exposure, with 75% of their tasks covered under the measure. Customer service representatives and data entry keyers also feature among the most exposed occupations. Financial analysts and other office-based roles show significant coverage as well.

At the lower end, about 30% of workers show zero exposure because their tasks rarely appear in AI usage data. These occupations include cooks, mechanics, lifeguards, bartenders and other roles that require physical work or in-person interaction.

Growth Projections and Worker Profile

The study compares its exposure measure with employment projections from the US Bureau of Labor Statistics for 2024 to 2034. It finds that occupations with higher observed exposure are projected to grow more slowly. For every 10 percentage point rise in exposure, projected job growth falls by 0.6 percentage points. The relationship is described as modest.

Workers in highly exposed professions also differ in profile from those in unexposed roles. They are more likely to be older, female, more educated and higher paid. On average, workers in the most exposed group earn 47% more than those in jobs with no exposure. Graduate degree holders account for 17.4% of the most exposed group, compared with 4.5% in the unexposed group.

Unemployment Trends Since 2022

To measure labour market impact, the researchers focus on unemployment. They compare workers in the top quartile of exposure with those in the bottom group using data from the Current Population Survey.

Since late 2022, they find no systematic rise in unemployment among highly exposed workers. The difference in unemployment rates between the most and least exposed groups remains small and statistically insignificant. The findings suggest that AI has not yet caused visible job losses at a broad level.

Signs of Slower Hiring for Young Workers

The study pays special attention to workers aged 22 to 25. While their unemployment rate in exposed occupations remains stable, hiring patterns show change. Monthly job entry into highly exposed occupations has declined by about half a percentage point.

On average, job-finding rates in exposed roles are 14% lower than in 2022. The result is described as just barely statistically significant. No similar decline appears for workers older than 25.

The researchers caution that slower hiring does not automatically mean displacement. Young workers may remain in existing jobs, shift to other occupations or return to education. Survey-based measures of job transitions may also include reporting errors.

The report concludes that there is limited evidence so far that AI has altered overall employment levels. However, the framework aims to track changes over time and identify early signs of disruption. Future updates will incorporate new usage data and revised task capability estimates to better understand how AI may reshape the labour market.