AI in Hiring and Management: The Fair Work Risk (2026)

So you buy this off-the-shelf AI tool to sort through resumes, right? You're just hoping to save yourself, I don't know, maybe 10 hours a week. Which makes total sense. Exactly.

But then a month later, you are standing in front of a Fair Work Tribunal for discrimination, and the law actually says you are guilty until proven innocent. Today we're doing a deep dive into a massive legal risk hidden in a new brief from Coterie Labs. Yeah, and look, this isn't just some abstract issue for multinational corporations. I mean, the latest data shows 60% of small businesses, so we're talking places with 5 to 19 staff.

Right. Like your local Sparky crew or an accounting firm. They are actually using AI right now to handle rostering or hiring. Okay, let's unpack this.

Because buying a sleek piece of software to screen apprentices, well, it feels like a really smart move for a time-poor business owner. Oh, of course it does. But thinking about it, it's actually a bit like hiring a dodgy subbie. Like if that subbie completely botches the wiring on a commercial build, you can't just point the finger at them when the building inspector shows up.

No, no, you can't. Because it's your site. You are on the hook. Spot on.

And if that dodgy subbie illegally used asbestos, you wouldn't just tell the inspector, well, I didn't tell him to use asbestos. Yeah, they wouldn't care. Exactly. And that is exactly how the Australian Human Rights Commission, the AHRC, views your automated HR tools.

The employer answers for the automated decisions, not the vendor who built the software. Right. Because under Section 361 of the Fair Work Act, you face what is called a reverse onus of proof. Hold on, a reverse onus?

So if an applicant claims my new $50 a month AI tool, knock them back for a discriminatory reason, the law just presumes they are right. Yeah, pretty much. How is a local builder supposed to prove the inner workings of a black box algorithm that they didn't even write? Well, it is nearly impossible.

I mean, you have to prove the algorithm didn't discriminate. Wow. And if you are relying purely on a machine that cannot explain its own reasoning, you know, you're going to have an incredibly hard time overcoming that reverse burden in court. But wait, if I never check a box in the software that says, you know, filter out candidates by race or gender, how does discrimination even happen?

I mean, the machine just reads the text on the resume. Yeah. So the trap here is that the AI doesn't actually need you to give it a specific race or gender filter to be biased. Really?

Yeah. It creates its own filters based on patterns and past data. It's a mechanism called proxy discrimination. Okay.

Proxy discrimination. What does that actually look like? Well, let's say the AI analyzes the last 10 years of your hiring data, right, to figure out what a successful employee looks like. It might notice that employees from a specific postcode tend to quit sooner.

Oh, because maybe that postcode is further away. So the commute burns them out. Right. Or that postcode happens to be a primarily immigrant neighborhood.

The AI doesn't know about the commute or the demographics at all. It just sees the raw data. Exactly. It just sees the data pattern and starts blanket rejecting everyone from that postcode to, you know, optimize retention.

So it's basically using the postcode as a proxy for ethnicity. Man, that is terrifying. Or I guess a gap in a resume. Like a human might see a two-year gap and think, oh, maternity leave.

Yep. Spot on. But the AI just calculates, well, gap equals higher turnover risk. Yeah.

And then automatically dumps the resumes of working mothers. Yeah. It just repeats and scales whatever historical biases were in the data it was fed. It doesn't know any better.

So what does this all actually mean for your business today? Like, how do we fix this? Well, to protect your business, the AHRC's rule of thumb is basically automate the admin, not the accountability. Automate admin, not accountability.

Got it. Practically, this means assigning an actual human to sign off on every single AI-driven rejection. So keeping a human in the loop, basically. The software can do the initial sorting, but a person has to hit the final no thanks button and actually verify the reason.

Yeah, exactly. And second, you also need to verify that your software vendors train their tools on Australian data. Right. Not overseas cohorts where the patterns might be completely different.

Yeah, because that skews everything. Third, you want to build an AI register. Just a simple spreadsheet listing every single tool that touches your rostering or hiring. That sounds like something you really want in place before December 10th, 2026.

The Privacy Act is changing then, right? Yeah, it's a hard deadline for transparency. You will be legally required to disclose when automated decisions significantly affect people. Like whether they get an interview or get rostered off.

Exactly. If an AI is making a call that impacts someone's livelihood, they have a legal right to know how that decision was made. Which leaves you with one massive question to chew on as you get back to the site or the office today. Mm-hmm.

You might know exactly what standalone hiring software you just bought. Mm-hmm. But what about the new AI features quietly being baked into the CRM or HR platforms you have used for years? Are they already acting like that dodgy subby in the background, making illegal decisions on your behalf without you even realizing?

That was a field note brought to you by Coterie Labs. We help Australian businesses work out where AI actually pays and where it doesn't. The full write-up with every source and link is in the show notes. And if you're wondering where AI fits in your own business, there's a free two-minute scan at coderielabs.com.au.

New field note every weekday.