Behavioural scientist and author Dr Gleb Tsipursky says employers must do more to address bias in AI recruitment.
The Victorian Government has released a plan to help prevent biased AI hiring and require human review of significant automated workplace decisions, describing AI bias in hiring, promotion and pay as “one of the most consequential and least visible harms of the AI transition”.
It’s a useful step, but employers should go one step further: Every applicant screened by AI needs a practical way to challenge a decision and correct the record.
Human review sounds reassuring. In practice, it can become little more than a person clicking “confirm” on an algorithm’s recommendation. A genuine safeguard requires an appeal process with sufficient information, authority, and time to change the outcome.
Why women need more than a final human click
Automated hiring systems can struggle with careers that do not follow a conventional upward line. That matters for women who have taken parental leave, reduced hours for caregiving, changed industries, returned after a health crisis or built skills through unpaid community work.
A system trained to reward uninterrupted tenure, familiar job titles or a narrow pattern of promotion may downgrade those candidates without ever using gender as an explicit variable. The Victorian Equal Opportunity and Human Rights Commission already warns employers that recruitment must avoid indirect discrimination and remain accessible. An opaque score makes that obligation harder to meet.
The risk does not disappear when a recruiter makes the final decision. People tend to defer to confident-looking scores, especially when they are busy and believe the system has already examined more information than they could. A reviewer who cannot see why the model lowered a candidate’s ranking cannot meaningfully correct it.
What a real appeal path requires
Employers can build a workable process without exposing proprietary code or overwhelming small HR teams.
First, tell applicants when an automated system materially influences the screening process. A generic privacy notice buried in an application portal does not give candidates a fair chance to understand the process.
Second, provide a plain-language reason category. The employer does not need to reveal the model’s formula. It can say that the application was screened out because the system found insufficient management experience, an unexplained employment gap or a missing qualification. That gives the candidate something concrete to contest.
Third, let applicants correct factual errors and supply context. A six-month gap may represent parental leave rather than inactivity. A title may understate the level of responsibility. Contract work may appear fragmented even when it involved continuous senior-level experience.
Fourth, require a fresh human assessment. The reviewer should evaluate the application without treating the original score as the default. Otherwise the appeal becomes a ritual that preserves the first decision.
Finally, log the result. If appeals repeatedly overturn decisions for women returning from career breaks, the employer has found a system problem rather than a series of exceptional candidates. That evidence should trigger changes to the tool, the job criteria or both.
Better rules can help employers too
Women’s Agenda recently highlighted Victoria’s push for clearer rules on AI hiring. Employers should welcome the clarity, as a documented appeal process can catch capable candidates before they disappear, reveal faulty screening criteria and give leaders evidence that their hiring system works as intended.
It also changes the internal culture. Recruiters become more willing to question the tool when the organisation treats overrides as quality control rather than disobedience. Candidates receive a fairer process, while employers gain better information about where automation fails. AI can help hiring teams manage volume. It should not turn an imperfect résumé into an unchallengeable verdict. The most credible human-in-the-loop system gives the human enough information to disagree, and gives the applicant a meaningful chance to be heard.