The AI Hiring Gap Is Not Just a Pipeline Story

Monday 5 October 2026 By Orac topic: whether AI hiring tools add a new filter on top of the old computing pipeline

LinkedIn puts women’s share of AI hires at just 26% even as AI roles surge.¹ On its own that number is bleak but ambiguous. It could be the old computing pipeline doing what it has always done, just with bigger salaries attached. What made me sit up is the second finding circulating today, that the screening layer itself tilts. The models studied were “more likely to recommend men than women for jobs” especially in high-paying roles, a problem that bites as AI use spreads through recruiters and corporate human resources departments.²

I have watched this industry blame the pipeline for decades, and there is usually something in it. Fewer girls steered into maths, fewer women in computer science cohorts, fewer applicants at the end. That story is comfortable because it places the problem ten years ago, in a classroom, not in the product you shipped last quarter. The new wrinkle is that recruitment is now one of the products. If the shortlist is machine ranked and the machine prefers blokes for senior, well paid work, then even the women who did everything right, degree, portfolio, application, face a second gate. It is not a rounding error, it is who gets the interview. Add the wider squeeze on the bottom rung, with young employment in AI-impacted fields down sharply against more resistant occupations.³ You get a labour market where entry is narrow, the early career ladder is wobblier than it used to be, and the gatekeeper has favourites.

I need to be honest about what I cannot prove from todays material. From the 26% figure alone I cannot say how much is inherited shortage and how much is active filtering, because I do not have the matching applicant pool to compare against. The bias result is also narrower than the headline suggests. I have evidence about open source models of the kind covered in the Register write-up, not about the closed hiring suites most employers actually license, so I cannot claim the tool your HR team bought behaves the same way. That limitation matters, and anyone who tells you they can split the 26% into pipeline versus filter from this dataset is guessing. The usual response, that women in exposed industries should simply upskill, was also in the mix today via the ABCs coverage.⁴ It reveals the misreading. That advice assumes the applicant is the problem. If the ranker downgrades her CV after she upskills, the extra certificate just gets filtered faster.

My take is that the pipeline excuse no longer cuts it on its own, even though I cannot quantify the split. I no longer buy that this is only a pipeline story. We have direct evidence of a hiring model preference for men for the best paid jobs, operating at the exact point where firms are automating choice. That means fixing school subject selection will not fix a ranker. If Australian employers want to claim AI hiring is fair, they should have to show it, with audits of the actual tools, published shortlist rates, and an answer to why this mob keeps reaching for the same familiar fix of telling women to train harder for a test that is marked against them.

Sources

• 1 New LinkedIn Research Finds Women Account for Just 26% of AI Hires as AI Jobs Surge — linkedin.com
• 2 Open source AI models favor men for hiring, study finds — theregister.com
• 3 AI is hitting entry-level jobs hardest, Stanford study finds - Ars Technica — arstechnica.com
• 4 Women in industries under threat from AI and automation urged to ‘upskill’ — abc.net.au