For years, the working assumption in AI policy circles was that caregiving jobs, mostly done by women, were safe from automation because they depend on human touch. New data out of Australia turns that assumption on its head. It is not caregiving jobs that are most at risk. It is office jobs, and those office jobs are also mostly done by women.
The Australian government agency Jobs and Skills Australia scored every occupation in the country on how exposed it is to AI automation. Clerical and administrative work, roles like secretaries, receptionists, bookkeepers, and payroll clerks, came out near the top of the risk list. That sector is more than 70 percent female and accounts for close to one in five jobs held by women in the country. Among the 20 occupations ranked most exposed to automation, 15 are female-dominated. Among the 20 occupations ranked least exposed, 17 are male-dominated, mostly hands-on trades like electrical work, where physical judgment on unpredictable job sites is hard for software to replace.
This is not a theoretical risk sitting years in the future. Job vacancy postings are already moving. Openings for personal assistants and bookkeepers fell by about 22 percent over the past year, and general clerk postings dropped by about 9 percent, while total job openings across the whole economy slipped by less than 2 percent. Trades are moving the opposite direction: postings for electricians rose by more than 14 percent over the same period.
Australia is not an outlier here. International labor researchers looking at the same question worldwide have reached the same conclusion: women are consistently overrepresented in the job categories that generative AI can most easily take over. In wealthier countries, the share of women's jobs sitting in the highest automation risk category runs nearly three times higher than the share for men. The reason is structural rather than about skill or effort. Clerical work tends to be routine, rules-based, and already done on a computer, which is exactly the kind of work current AI tools are good at handling.
There is a second layer to this problem. Women make up only around 30 percent of the people working in AI itself, so the group facing the most job disruption has the least influence over how the technology gets designed and deployed. That gap matters because decisions about which tasks get automated, and how quickly, are being made largely without input from the workers most affected.
None of this means retraining programs are a simple fix. Research on past waves of worker retraining shows mixed results at best, so treating "reskill them" as an automatic solution understates the difficulty. What actually helps is more specific: mapping the coordination, communication, and judgment skills clerical workers already have onto new roles, and doing that mapping before layoffs happen rather than after.
For any business with a sizable administrative or clerical workforce, the practical lesson is timing. The companies that get ahead of this will treat workforce planning as an ongoing task tied to real hiring and workload data, not a one-time reaction to a bad quarter.