Press kit

The Task Exposure Index rates 923 occupations task by task on how much of their work current AI systems can produce, and what stands in the way when they can. It works from the 2,087 standardised work activities behind 18,838 O*NET tasks, each rated on six published dimensions, with inter-rater agreement measured and reported. It is re-scored every quarter against a stated capability date, and old releases are never edited. Release v2026.Q3 is rated against 2026-09-15.

• The lowest paid fifth of US jobs is the least exposed. Weighted by employment, the bottom pay quintile has 23.9% of its work exposed, the top two quintiles 39.2% and 37.4%. The analysis.
• Jobs that need a bachelor’s degree are more exposed than jobs that need a doctorate or licence. O*NET Job Zone 4 averages 41.1% exposed, Zone 5 35.6%, Zone 2 16.4%. The analysis.
• Context, not hands, is what holds AI back most. Across all 18,838 tasks, context accounts for 24.7% of the friction and embodiment for 21.6%, and context is the one that erodes. The analysis.
• Paralegals are more exposed than lawyers (33.6% against 22.0%). The analysis.
• Scheduling is the most exposed kind of work: the 14 scheduling activities average 80.5% exposed. The analysis.

The full ranking is at rankings and as a CSV.

Exposure is not displacement. A high score says current AI systems can produce the work, not that an employer will stop paying a person to do it. Please do not report a score as “X% of jobs will be lost”. Why.

Free to quote and chart under CC BY 4.0. Credit “Task Exposure Index” with the release, v2026.Q3, and link to https://taskexposure.org. Every job page has a share image, an embeddable widget and a citation block. For a figure we do not publish, a custom cut or the next release under embargo, write to press@taskexposure.org.

The research desk at A.I.T. Multiverse Consulting Ltd, Nicosia, Cyprus. No AI company, staffing firm, training provider or government body funds the index. About · Methodology · Funding and disclosure.