{
  "@context": {
    "@vocab": "http://schema.org/",
    "exposure": "http://schema.org/occupationalExposure",
    "BLS": "https://www.bls.gov/",
    "O_NET": "https://www.onetonline.org/",
    "AnthropicEconomicIndex": "https://huggingface.co/datasets/Anthropic/EconomicIndex",
    "EloundouEtAl2023": "https://arxiv.org/abs/2303.10130",
    "CPS": "https://www.bls.gov/cps/",
    "CausalInference": "http://schema.org/StatisticalPopulation",
    "AIExposureMeasure": "http://schema.org/PropertyValue",
    "LLM": "http://schema.org/SoftwareApplication",
    "API": "http://schema.org/SoftwareApplication",
    "HowToStep": "http://schema.org/HowToStep",
    "DefinedTerm": "http://schema.org/DefinedTerm",
    "Question": "http://schema.org/Question",
    "Answer": "http://schema.org/Answer",
    "Article": "http://schema.org/Article",
    "Author": "http://schema.org/Person",
    "Publisher": "http://schema.org/Organization",
    "Dataset": "http://schema.org/Dataset",
    "Table": "http://schema.org/Table",
    "ImageObject": "http://schema.org/ImageObject"
  },
  "@type": "Article",
  "headline": "Labor market impacts of AI: A new measure and early evidence",
  "author": [
    {
      "@type": "Person",
      "name": "Maxim Massenkoff"
    },
    {
      "@type": "Person",
      "name": "Peter McCrory"
    }
  ],
  "datePublished": "2026-03-05",
  "publisher": {
    "@type": "Organization",
    "name": "Anthropic",
    "url": "https://www.anthropic.com"
  },
  "abstract": "This article introduces a new measure of AI displacement risk called observed exposure, combining theoretical LLM capability and real-world usage data, and presents early evidence on labor market impacts of AI.",
  "articleBody": "The article presents a novel framework for measuring AI's impact on labor markets by combining theoretical AI task capability with observed real-world usage data to create an observed exposure metric. It finds that AI adoption is far from theoretical limits, with occupations like computer programmers and customer service representatives most exposed. Early evidence shows no systematic increase in unemployment for highly exposed workers but suggests slowed hiring among younger workers in exposed occupations.",
  "hasPart": [
    {
      "@type": "DefinedTermSet",
      "name": "Key Defined Terms",
      "description": "Important terms defined in the article related to AI labor market impacts.",
      "hasDefinedTerm": [
        {
          "@type": "DefinedTerm",
          "name": "Observed Exposure",
          "description": "A measure quantifying the share of tasks that LLMs could theoretically speed up and are actually seeing automated usage in professional settings."
        },
        {
          "@type": "DefinedTerm",
          "name": "Theoretical AI Capability",
          "description": "The potential of LLMs to perform tasks at least twice as fast, as estimated by Eloundou et al. (2023)."
        },
        {
          "@type": "DefinedTerm",
          "name": "LLM (Large Language Model)",
          "description": "A type of AI model capable of understanding and generating human-like text, used to assess task automation potential."
        },
        {
          "@type": "DefinedTerm",
          "name": "Task Coverage",
          "description": "The fraction of job tasks that AI can perform or assist with, weighted by time spent on each task."
        },
        {
          "@type": "DefinedTerm",
          "name": "Automated Use",
          "description": "AI usage where tasks are fully automated without human augmentation."
        },
        {
          "@type": "DefinedTerm",
          "name": "Augmentative Use",
          "description": "AI usage where AI assists humans but does not fully automate tasks."
        },
        {
          "@type": "DefinedTerm",
          "name": "BLS Projected Employment Growth",
          "description": "Employment growth forecasts from the US Bureau of Labor Statistics for occupations from 2024 to 2034."
        },
        {
          "@type": "DefinedTerm",
          "name": "Current Population Survey (CPS)",
          "description": "A US survey used to analyze unemployment and labor market outcomes."
        },
        {
          "@type": "DefinedTerm",
          "name": "Difference-in-Differences Framework",
          "description": "A statistical method used to estimate causal effects by comparing changes over time between treated and control groups."
        },
        {
          "@type": "DefinedTerm",
          "name": "Economic Index",
          "description": "Anthropic's dataset measuring AI usage across tasks and occupations."
        }
      ]
    },
    {
      "@type": "HowTo",
      "name": "Measuring AI Exposure in Labor Markets",
      "description": "Steps to measure AI exposure combining theoretical capability and observed usage.",
      "step": [
        {
          "@type": "HowToStep",
          "position": 1,
          "name": "Collect Task Data",
          "text": "Use the O*NET database to enumerate tasks associated with around 800 US occupations."
        },
        {
          "@type": "HowToStep",
          "position": 2,
          "name": "Measure Theoretical AI Capability",
          "text": "Use Eloundou et al. (2023) task-level exposure estimates to score tasks by whether an LLM can perform them at least twice as fast."
        },
        {
          "@type": "HowToStep",
          "position": 3,
          "name": "Gather Real-World Usage Data",
          "text": "Collect AI usage data from the Anthropic Economic Index, measuring actual automated and augmentative AI use."
        },
        {
          "@type": "HowToStep",
          "position": 4,
          "name": "Calculate Observed Exposure",
          "text": "Combine theoretical capability and observed usage, weighting fully automated tasks more heavily, and aggregate to occupation level weighted by time spent on tasks."
        },
        {
          "@type": "HowToStep",
          "position": 5,
          "name": "Analyze Labor Market Outcomes",
          "text": "Compare observed exposure to BLS employment projections and CPS unemployment data to assess AI's labor market impact."
        }
      ]
    },
    {
      "@type": "Question",
      "name": "What is observed exposure in the context of AI labor market impacts?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Observed exposure quantifies the share of tasks that AI could theoretically automate and are actually seeing automated usage in professional settings, reflecting real-world AI adoption."
      }
    },
    {
      "@type": "Question",
      "name": "How does observed exposure differ from theoretical AI capability?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Theoretical AI capability measures the potential for AI to perform tasks, while observed exposure measures the actual usage of AI in performing those tasks in the real world."
      }
    },
    {
      "@type": "Question",
      "name": "Which occupations are most exposed to AI according to the study?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Computer programmers, customer service representatives, and data entry keyers are among the most exposed occupations, with observed exposure levels above 65%."
      }
    },
    {
      "@type": "Question",
      "name": "What is the relationship between AI exposure and projected job growth?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Occupations with higher observed AI exposure tend to have weaker projected employment growth according to BLS forecasts, with a 10 percentage point increase in exposure associated with a 0.6 percentage point drop in growth projection."
      }
    },
    {
      "@type": "Question",
      "name": "Has AI exposure led to increased unemployment so far?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "No systematic increase in unemployment has been observed for highly exposed workers since late 2022, though hiring of younger workers in exposed occupations has slowed."
      }
    },
    {
      "@type": "Question",
      "name": "How does AI exposure vary by worker demographics?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Workers in highly exposed occupations are more likely to be older, female, more educated, and higher-paid compared to those in low exposure occupations."
      }
    },
    {
      "@type": "Question",
      "name": "What data sources are used to measure AI exposure?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "The study uses O*NET task data, Anthropic Economic Index usage data, and task-level exposure estimates from Eloundou et al. (2023)."
      }
    },
    {
      "@type": "Question",
      "name": "What is the significance of the difference-in-differences framework in this study?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "It helps isolate the effect of AI exposure on unemployment by comparing changes over time between workers in highly exposed and unexposed occupations."
      }
    },
    {
      "@type": "Question",
      "name": "What early signals of AI impact on labor markets are identified?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "There is suggestive evidence of slowed hiring among young workers (ages 22-25) in highly exposed occupations, though unemployment rates remain flat."
      }
    },
    {
      "@type": "Question",
      "name": "Where can the observed exposure data be accessed?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Observed coverage data at the task and job level is available at https://huggingface.co/datasets/Anthropic/EconomicIndex."
      }
    },
    {
      "@type": "Table",
      "name": "Most Exposed Occupations",
      "description": "Top ten occupations ranked by observed AI exposure with leading automated tasks.",
      "url": "page 7",
      "about": [
        "Computer programmers (74.5%) - Write, update, and maintain software programs",
        "Customer service representatives (70.1%) - Confer with customers to provide info, take orders, handle complaints",
        "Data entry keyers (67.1%) - Read source documents and enter data into systems",
        "Medical record specialists (66.7%) - Compile, abstract, and code patient data",
        "Market research analysts and marketing specialists (64.8%) - Prepare reports of findings, illustrate data graphically",
        "Sales representatives (62.8%) - Contact customers to demonstrate products and solicit orders",
        "Financial and investment analysts (57.2%) - Inform investment decisions by analyzing financial info",
        "Software quality assurance analysts and testers (51.9%) - Modify software to correct errors or improve performance",
        "Information security analysts (48.6%) - Perform risk assessments and test data processing security",
        "Computer user support specialists (46.8%) - Answer user inquiries regarding software or hardware"
      ]
    },
    {
      "@type": "ImageObject",
      "name": "Theoretical capability and observed exposure by occupational category",
      "description": "Radar chart comparing theoretical AI coverage (blue) and observed AI coverage (red) across occupational categories.",
      "contentUrl": "page 6 image",
      "caption": "Figure 2 on page 6"
    },
    {
      "@type": "ImageObject",
      "name": "BLS projected employment growth vs. observed exposure",
      "description": "Scatterplot showing negative correlation between AI exposure and projected employment growth from 2024 to 2034.",
      "contentUrl": "page 8 image",
      "caption": "Figure 4 on page 8"
    },
    {
      "@type": "ImageObject",
      "name": "Unemployment trends for high vs. no AI exposure workers",
      "description": "Line charts showing unemployment rates and difference-in-differences estimates for workers in top quartile of AI exposure vs. no exposure from 2016 to 2025.",
      "contentUrl": "page 11 image",
      "caption": "Figure 6 on page 11"
    },
    {
      "@type": "ImageObject",
      "name": "New job starts among young workers by AI exposure",
      "description": "Line charts showing monthly job start rates for workers aged 22-25 in high vs. no AI exposure occupations, with difference-in-differences estimates.",
      "contentUrl": "page 13 image",
      "caption": "Figure 7 on page 13"
    }
  ],
  "citation": {
    "@type": "CreativeWork",
    "name": "Labor market impacts of AI: A new measure and early evidence",
    "author": [
      {
        "@type": "Person",
        "name": "Maxim Massenkoff"
      },
      {
        "@type": "Person",
        "name": "Peter McCrory"
      }
    ],
    "datePublished": "2026-03-05",
    "url": "https://www.anthropic.com/research/labor-market-impacts"
  },
  "mainEntity": [
    {
      "@type": "Question",
      "name": "How to measure AI exposure in occupations?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Combine task-level theoretical AI capability (Eloundou et al. 2023) with observed AI usage data (Anthropic Economic Index), weight automated use more heavily, and aggregate to occupation level using task time fractions."
      }
    },
    {
      "@type": "Question",
      "name": "What early labor market impacts of AI have been observed?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "No systematic increase in unemployment for highly exposed workers since late 2022, but suggestive evidence of slowed hiring among young workers in exposed occupations."
      }
    }
  ],
  "author": [
    {
      "@type": "Person",
      "name": "Maxim Massenkoff"
    },
    {
      "@type": "Person",
      "name": "Peter McCrory"
    }
  ],
  "publisher": {
    "@type": "Organization",
    "name": "Anthropic",
    "url": "https://www.anthropic.com"
  }
}