The short answer: task exposure describes what the technology could speed up. Job loss depends on whether anyone buys the tool, whether the law still requires a person, whether the output clears your quality bar, whether demand for the occupation grows anyway, and whether the exposed tasks were what you were being paid for. Each of those has already been the reason a confident automation prediction failed.
Two different objects
An exposure figure is a statement about the composition of a job. Under the Science rubric, 60% exposure means roughly 60% of the tasks, or of the hours, land on work that a language model or software built on one could do at least twice as fast at the same quality. A probability of job loss is a statement about an event happening to a person. They are measured in different units and they answer different questions.
The published research measures the first. It does not measure the second, and the papers say so. What happens between a measurement and an outcome is filled with things no exposure rating touches.
The five things in the gap
Somebody has to buy the tool. Task exposure is a property of the work. Adoption is a property of an organisation, and the distance between them is measured in procurement cycles, security reviews, training, and years. Plenty of exposed tasks are still done by hand in firms that have not made a decision yet.
The law may require a person regardless. An audit opinion, a prescription, an engineering stamp, a court filing: each carries a named person whose licence is on the line. No efficiency gain changes that. In regulated fields the exposed task can be fully automatable and still legally require a human to perform or sign it.
Output has to clear your quality bar, not a demo's. The ratings ask whether output holds up in general. They cannot ask whether it holds up for your clients, at the standard your firm is held to, with the consequences you carry when it is wrong. That is an empirical question about your workplace, and no general rating answers it.
Demand for the occupation may grow anyway. Software development carries some of the highest measured task exposure in the research, and BLS projects employment up 15% from 2024 to 2034. An occupation can shed tasks and add headcount at the same time if demand grows faster than the tasks shrink.
The absorbed tasks may not have been what anyone was paying for. A great deal of necessary work is not the reason a role exists. If the exposed hours are the ones your employer tolerated rather than valued, removing them changes your week without changing your value.
Each of those five has already been the reason a confident automation prediction failed. Frey and Osborne's 2013 estimate put 47% of US employment at high risk over a decade or two. Most of that window has passed.
Why the swap keeps happening
A probability is a better headline. It is easier to share and easier to sell, and it sounds like a prediction someone stood behind. An exposure share sounds like a measurement, which is what it is. The swap usually happens between the study and the headline, not inside the study, and the researchers are rarely the ones making it.
How to use an exposure number
Read it as a description of your week, not a verdict on your career. A high number tells you which hours are where the technology can already reach. It does not tell you what your employer will do, what your regulator will require, or what the market for your occupation looks like in five years. Then ask which of the five gaps is doing the most work in your own situation. For most people one of them dominates, and that one is the thing worth watching. The free AI Job Risk Score reports the measurement and stops there, on purpose.
Frequently asked questions
Does a 60% AI exposure score mean a 60% chance of losing my job?
No. Exposure measures the share of your tasks or hours that AI could speed up. A probability of job loss would have to account for adoption, regulation, quality standards, demand for the occupation, and what your employer actually values, none of which the exposure rating measures.
Has any study predicted the probability that AI eliminates a specific job?
Not in the peer-reviewed exposure research. Eloundou et al. (2024) and similar work measure task exposure and state that it should not be read as a displacement forecast.
Can an occupation be highly exposed and still grow?
Yes. Software developers carry some of the highest task exposure in the research, and BLS projects employment growth of about 15% from 2024 to 2034.
Get your free AI Job Risk Score. Tell us your job title and how you actually spend your time, and we will show you which of your tasks are exposed today. Free. 60 seconds.
Get my AI risk now