Victoria, the Australian state that includes Melbourne, has announced plans to regulate how AI is used in hiring and workplace monitoring. The proposals would require a human to be involved in every AI-assisted hiring decision, ban tools that try to read candidates' emotions from their faces or voices, and give employees the right to know if and how they are being monitored.
This is not a niche issue. Around 83% of companies plan to use AI to screen resumes, and about 29% want AI to handle the entire interview process end-to-end. The pitch from software vendors is straightforward: faster, cheaper, more consistent. Companies report that AI screening can cut hiring time by up to 50% and reduce cost per hire by around 30%. When you are receiving thousands of applications for one role, that efficiency is hard to pass up.
The problem is that the same systems carry hidden sorting errors. AI hiring tools learn patterns from past hiring data. If past hires were mostly white, mostly male, or mostly from a narrow range of universities, the system learns to prefer those patterns going forward. Research has found that AI resume screening tools favor white-associated names a large majority of the time and favor male-associated names in most cases. In a 2025 class-action lawsuit in the United States, one major software platform was accused of systematically filtering out applicants based on race, age, and disability.
The problem extends to video interviews. Tools that claim to assess confidence, enthusiasm, or cultural fit by analyzing facial expressions, voice tone, or body language have no solid scientific foundation. The EU took this seriously enough to ban emotion-recognition AI at work outright from February 2025. Under those rules, using such tools in recruitment can now draw fines of up to 35 million euros or 7% of global turnover.
Australia's federal government moved in the opposite direction. Its December 2025 national AI plan dropped a previous proposal for mandatory guardrails on high-risk AI and instead asked regulators to identify gaps in existing law. That left states like Victoria and New South Wales to act on their own. NSW passed its own AI workplace safety law in February 2026. Victoria's proposals are contingent on Labor winning the November 2026 state election.
Here is what this means for any employer, regardless of geography. Under current Australian law, if an AI hiring tool filters out candidates based on race, age, gender, or disability, the employer is liable, not the software vendor. Courts have been consistent on this: buying a biased tool is not a defense. The same logic applies under existing anti-discrimination law in most comparable countries.
There is also a practical talent cost that tends to get overlooked. A US survey found that 66% of job seekers say they would refuse to apply for jobs where AI plays a major role in hiring. If your screening tool quietly drives away half the candidate pool before they even apply, you are not saving time. You are reducing your options.
The clearest action for any employer using AI to filter applications is to ask the vendor directly: how was this tool tested for bias, and against which groups? If the vendor cannot show documented testing and audit results, that is the answer. Most tools on the market cannot. The Victorian proposals would formalize what good employers should already be doing: keeping a human in the loop, auditing for patterns of exclusion, and not using tools that have been scientifically discredited.