AI giant Anthropic compared observed AI exposure and theoretical AI capability in the labour market. The analysis shows that AI is still far from reaching its theoretical potential.
The impact of Artificial Intelligence on jobs has become one of the defining debates of the moment, with international organisations, academics and hiring companies regularly publishing projections on which professions are most at risk.
ADVERTISEMENT
ADVERTISEMENT
A new entry in that crowded field now comes from one of the AI giants itself.
Anthropic, the company behind Claude, has published a report titled Labour Market Impacts of AI: A New Measure and Early Evidence, based on its own real-world usage data.
Theoretical capability vs observed exposure
The report introduces a new measure called "observed exposure" — designed to quantify not just which tasks large language models could theoretically speed up, but which are already being automated in practice.
The distinction matters: theoretical capability reflects what AI could do while observed exposure reflects what it is actually doing.
Highest theoretical AI coverage: Computer, math, business and finance
The theoretical AI coverage exceeds 80% in several occupation groups among the 22 analysed. Computer and math, as well as business and finance occupations have the highest theoretical AI coverage both at 94.3%.
Other groups with theoretical capability above 80% include management (91.3%), office and administrative support (90%), legal (89%), architecture and engineering (84.8%), and arts and media (83.7%).
In five additional occupational groups, scope for LLM penetration exceeds 50%.
These include life and social sciences (77%), sales (62%), education and library occupations (61.7%), healthcare practitioners (59.9%), and social services (50.5%).
The red area above, based on data from the Anthropic Economic Index, shows how people use Claude in professional settings.
“As capabilities advance, adoption spreads, and deployment deepens, the red area will grow to cover the blue. There is a large uncovered area too; many tasks, of course, remain beyond AI’s reach—from physical agricultural work like pruning trees and operating farm machinery to legal tasks like representing clients in court,” the report said.
Lowest ‘potential’ include transportation, agriculture and food
Theoretical AI coverage is lowest in ground maintenance where only 3.9% of jobs in this group are theoretically open to AI usage.
Transportation (12.1%), agriculture (15.7%), food and serving (16.9%), construction (16.9%), personal care (18.2%), installation and repair (18.4%), and production (19%) also have significantly lower theoretical AI coverage, all below 20%.
This suggests there may be less potential space for AI use in these sectors.
Theoretical AI coverage is also lower in healthcare support (28.5%) and protective services (31.6%).
Highest observed exposure: Computer, math, office and admin
The more important question is to what extent theoretical capability has turned into observed exposure, showing AI displacement risk.
Computer and math occupations have the highest observed AI coverage at 35.8%, followed closely by office and administrative roles (34.3%).
Business and finance (28.4%) and sales (26.9%) are also close to these levels.
Legal (20.4%), arts and media (19.2%), and education and library occupations (18.2%) also have relatively high observed AI exposure at around 20%.
Observed exposure as a share of theoretical AI capability
The ratio of observed exposure to theoretical capability shows to what extent this potential is already being used.
Sales tops the list at 43% (27% vs 62%), followed by office and administrative jobs (38%) and computer and math occupations (38%).
Observed exposure as a share of theoretical AI capability is 30% in business and finance, and in education and library occupations.
While architecture and engineering have very high theoretical capability (85%), the ratio is just 5%.
Most exposed occupations: Computer programmers and customer service representatives
Among individual occupations, computer programmers show the highest observed AI exposure at 74.5%.
Customer service representatives (70.1%), data entry keyers (67.1%), and medical record specialists (66.7%) also rank among the most exposed.
Market research analysts and marketing specialists follow at 64.8%, along with sales representatives in wholesale and manufacturing, except technical and scientific products (62.8%).
The data also sheds light on who is most exposed. Workers in the highest-risk professions tend to be older, more educated, better-paid and more likely to be women.
Yet so far, at least, exposure has not translated into unemployment.
The report found no systematic increase in joblessness among workers in heavily exposed occupations since late 2022, though it did find suggestive evidence that hiring of younger workers has slowed in those same fields — a detail worth watching.
