Annual report

The 2026 Professional AI Exposure Index

Model v1.2 · indices last updated 18 August 2026 · O*NET 30.3

Every figure below is a task exposure index: how much of a typical working week for a job title is made of activities current AI systems can already perform end to end, weighted by the time each activity takes. It is an exposure index, not a prediction — it says nothing about what happens to any job, and it has no opinion about dates. And every number is the average for a job title, not a measurement of any particular person’s week; the difference between the two is usually larger than the difference between job titles.

occupations scored: 923
median exposure index: 45
in the high band: 226
in the low band: 281

The most exposed occupations

The highest task exposure indices of the 923 occupations scored. Each row links to the full decomposition.

The least exposed occupations

The lowest indices — the occupations whose weeks are built most heavily from work current systems handle worst.

Every field, ranked

The average index across each O*NET major group, most exposed first.

How these numbers are made, and what they cannot say

O*NET 30.3 decomposes each occupation into its work activities and how much of a typical week each takes. The exposure figure for each activity is ours — O*NET publishes no such number — recorded under model v1.2 and re-examined quarterly against evidence of what AI systems demonstrably shipped, with every round on the public changelog. The full method, weights and limitations are on the methodology page. What the index cannot see: cost, regulation, liability and inertia — automatable is not automated.

Your own week is what the assessment measures — two people with the same job title get different numbers, because they spend their weeks differently.