Anthropic finds signs of slower hiring among young workers in AI-exposed jobs

Anthropic introduced “observed exposure” to assess AI displacement risk by combining what large language models could theoretically do with data on how AI is actually used.[1] The measure gives more weight to uses that automate work than to those that augment it.[1] Anthropic says the framework aggregates measures of theoretical capability and real-world usage to the occupation level.[1] That design is intended to distinguish potential capability from adoption in practice; Anthropic also says actual AI coverage remains only a fraction of what is theoretically feasible.[1]

Anthropic says the US Bureau of Labor Statistics projects lower growth through 2034 for occupations with higher observed AI exposure.[1] Its analysis also found that workers in the most exposed professions are more likely to be older, female, more educated and higher-paid.[1] The exposure measure is designed to help identify jobs that may be vulnerable before displacement becomes visible, according to Anthropic.[1]

Anthropic found suggestive evidence that hiring of younger workers has slowed in occupations exposed to AI.[1] But it found no systematic increase in unemployment among highly exposed workers since late 2022, and says its analysis of early data found limited evidence that AI has affected employment so far.[1] The hiring signal therefore does not establish that AI caused a general increase in unemployment. Anthropic says labor-market effects may not show up immediately in aggregate unemployment data.[1]

Anthropic notes that trade policy and the business cycle can complicate interpretation of labor-market trends.[1] It also says research on industrial robots’ employment effects reaches opposing conclusions, while the scale of job losses attributed to the China trade shock remains debated.[1] Government occupational growth forecasts, the company says, have added little predictive value beyond extending past trends linearly.[1] Anthropic plans to revisit its employment analyses periodically, so the current findings are an early assessment rather than a final account of AI’s labor-market effects.[1]