It can be, and the more useful finding is that hardly anyone in Australia checks. Dr Natalie Sheard's study of employers using AI hiring systems, published in the Journal of Law and Society in 2025, found these tools running with little testing for discriminatory outcomes and little disclosure to the people being filtered. An AI hiring system is any tool that ranks, scores or rejects applicants before a human reads them. A biased one and a working one look identical from where you sit.
Treasury put the stakes in writing on 21 September. The 2026 Intergenerational Report runs the budget to 2065-66, and the Treasurer's media release calls artificial intelligence the most dramatic change since the 2023 report, with income per capita 55 per cent higher by then. None of that arrives because you bought a tool. It arrives because the work changed.
Most Australian businesses of ten to fifty people have already bought AI. They bought it for the resume pile, the one place it changes nothing about how the work gets done.
Outstaffer sells an AI recruiter with a human validation step, so read this as disclosure rather than pitch.
The only Australian audit of AI resume screening at scale
Services Australia ran a bulk recruitment on what the Merit Protection Commissioner called AI-assisted and automated selection techniques: psychometric testing, questionnaires and self-recorded video, with no selection panel to cull or interview anyone. Promotion Review Committees overturned 11 of 115 promotion decisions.
The 2021-22 annual report found the process "did not always meet the key objective of selecting the most meritorious candidates", because the stepped online assessments did not adequately test a candidate's qualities against the qualities the duties genuinely required.
One in ten, near enough. That agency had a review mechanism, a union and a reporting obligation. Your shortlist has none of those.
Would you find out if your screen was wrong?
Almost certainly not. The people it filtered out never come back to tell you. Sheard's participants described systems chosen on vendor claims and left unaudited, so the loop that would expose a bad filter never closes. Discrimination law applies to recruitment whoever does the sorting, so get advice on your own process.
What biased resume screening actually costs you
Run that rate against a thirty-person business filling six roles a year and you are making a few dozen ranking calls with roughly one in ten reversible on review. Outstaffer publishes the thing it argues against, a $10,000 to $30,000 agency tax per hire, and one role you run twice sits inside that range on its own.
What does putting the judgement back cost?
Less than the fee does. Virtual Coworker publishes a recruitment assistant at $12 an hour AUD, an "Inclusive Hourly Rate, No Extra Fees", at 20 to 40 dedicated hours per week. Twenty hours a week for a year is $12,480: one agency placement at the bottom of that range, spent instead on someone reading what your filter ranked. On fees generally, the fee you can see is the only one you can argue with.
The strongest argument against all of this
Humans are not a clean baseline. Outstaffer's own published figures put reports of gender-based discrimination at 3.3 per cent for its AI interviewer against 6.0 per cent for human interviewers, a real number pointing the other way. It is also a number about reports rather than outcomes, and it comes from the vendor. A tested machine probably does beat a tired human at four in the afternoon. An untested one is your existing bias with a faster clock and no audit trail, a distinction we have made before about the point where speed stops being the useful measure.
Related reading
Treasury is not asking whether you adopted AI. It assumed you did. The productivity it banks on comes from redesigned work, and a filter bolted onto an unchanged process is not that. My position: if you cannot say who tested your screen, when, and against what, you do not have an AI advantage. You have an unexamined liability with a subscription attached.
So ask it about your own pile. If nobody knows, run one role through a process with a named human gate and compare that shortlist against the one your filter would have produced. What would you do if they disagreed?
More from
Global Hiring
category
%201.svg)
