Which Jobs Are Most Exposed to AI? Data from 800 Occupations

The Top 10 Most Exposed Occupations

In Part 1, we explained how Anthropic's "observed exposure" metric works—combining theoretical AI capability with real-world usage data. Now let's look at which specific occupations land at the top of that ranking.

The research analyzed approximately 800 occupations from the O*NET database, measuring what percentage of each job's tasks are actually being performed by AI in professional settings. Here are the ten most exposed:

Computer Programmers lead at 75% coverage. This aligns with other data showing coding is one of the most common professional use cases for AI. But notice what "75% coverage" means: three-quarters of the tasks in the occupation are being performed with AI in observed usage. The remaining 25% likely involves tasks requiring physical presence, complex stakeholder management, or highly specialized domain knowledge that hasn't yet migrated to AI workflows.

The 30%: Workers with Zero Exposure

At the other end of the spectrum, 30% of all workers have zero observed AI coverage. Their tasks appeared too infrequently in the usage data to meet the minimum threshold. These jobs include:

The pattern is clear: jobs requiring physical presence, manual dexterity, or direct human interaction in specific physical environments remain largely untouched by AI. This is consistent with AI's current strengths (language, analysis, code) and weaknesses (physical manipulation, real-world sensing).

The Surprise: Who AI-Exposed Workers Actually Are

Perhaps the most striking finding in the paper is who the most AI-exposed workers are. Using Current Population Survey data from the three months before ChatGPT launched (August–October 2022), the researchers compared demographics of the top quartile of exposed workers versus those with zero exposure.

This challenges the popular narrative that AI primarily threatens low-wage, low-skill workers. The data tells the opposite story: the most AI-exposed workers are more educated, higher paid, older, and more likely to be female compared to unexposed workers.

This makes sense when you consider what AI does well. LLMs excel at language, analysis, and information processing—the core of knowledge work. They cannot (yet) cook meals, repair engines, or guard swimming pools. The irony: the workers who spent the most on education and built the highest-paying careers are the ones most theoretically vulnerable to AI.

BLS Projections Align With AI Exposure

The Bureau of Labor Statistics publishes employment projections every two years. The latest set, published in 2025, covers predicted changes from 2024 to 2034. The researchers found a meaningful correlation:

While a 0.6 percentage point reduction per 10 points of coverage may sound small, consider the cumulative effect. An occupation with 75% coverage (like Computer Programmers) would see projected growth reduced by about 4.5 percentage points compared to an unexposed job. Over a decade, that's the difference between a growing career field and a shrinking one.

The Automation vs. Augmentation Question

A critical nuance in the research is the distinction between automated and augmentative AI use:

The observed exposure metric gives automated use full weight and augmentative use half weight. This means Customer Service Representatives rank highly not just because people use AI to help with customer service, but because AI is increasingly replacing the human in the loop via automated chatbots and API integrations.

For career planning, this distinction matters enormously. If AI is primarily augmenting your work (making you faster), your job may be safer—even enhanced. If AI is automating your work (replacing you in the loop), the trajectory points toward displacement.

What This Means by Career Stage

The Bottom Line

Anthropic's data gives us the clearest picture yet of which jobs AI is actually affecting—not theoretically, but in practice. The key takeaways:

• Programmers, customer service reps, and data entry workers are the most exposed today
• 30% of workers have essentially zero AI exposure, concentrated in physical, hands-on roles
• The most exposed workers are higher paid, more educated, older, and more likely female than average
• BLS growth projections are weaker for more AI-exposed occupations, validating the measure
• The distinction between automation and augmentation is critical for understanding risk

In Part 3, we tackle the ultimate question: is AI actually displacing workers right now? The employment data tells a nuanced story that neither the optimists nor pessimists will fully love.