Fair Work Billed a Worker $1,230 for Careless AI Use

Fair Work just made someone pay for careless AI. It almost never does that.

The Fair Work Commission is a 'bear your own costs' jurisdiction — win or lose, each side usually pays its own way. So when it ordered a sacked ALDI worker to pay $1,230 toward the supermarket's legal bill, that was the rare exception, and the reason was his use of AI. He'd run an unfair-dismissal claim built on AI-generated submissions that never grappled with the one fact that sank him. From 20 October 2026, the rules governing AI in every Fair Work matter become mandatory — and this case is the clearest signal yet of how the Commission will treat the people who ignore them.

What happened in Khan v ALDI

Sadnan Khan was dismissed three days before he reached the six-month minimum employment period — the threshold you must clear to bring an unfair-dismissal claim at all. He lodged an AI-generated application that same evening. The problem: eligibility turns on the date you're notified of the dismissal, and his AI-drafted submissions kept arguing the wrong date. Deputy President Easton warned him more than once that his case had a 'very strong likelihood' of failing. Khan pressed on, filed further AI-drafted material making the same error, then conceded at the hearing that he was never eligible.

In Khan v Aldi Pty Ltd [2026] FWC 3144, the Deputy President ordered him to pay $1,230 — the costs ALDI ran up after the warnings — under section 400A of the Fair Work Act, which allows a costs order where a party's unreasonable act or omission causes the other side to incur costs. Easton found that failing to critically check the AI's output against the Commission's own guidance was exactly that. The Commission's Executive Director has told a Senate committee that early research suggests 40 to 50 per cent of applicants now use AI to make a claim. This is the first costs order of its kind. It won't be the last.

From 20 October, the rules bind you too

The costs order used powers the Commission already had. What changes this month is a standing rulebook. From 20 October 2026, anyone using generative AI to prepare a document for the Commission must do three things: disclose that AI was used and how, verify that every fact, case citation and quotation is accurate and genuinely exists, and — for a witness statement — make sure the words are the witness's own, not the model's. It applies to all individual rights matters, unfair dismissal and general protections alike, and to self-represented parties and lawyers equally. Professional representatives, which the Commission takes to include HR advisers as well as lawyers, carry a higher bar: they must hyperlink every case they cite. We set out the full disclose-and-verify checklist here.

Get it wrong and the Commission can give your documents less weight, disregard them entirely, dismiss your case, or — as Khan found — order you to pay the other side's costs.

Why this cuts both ways for employers

If you run a business and someone you sacked files a claim, you're usually the respondent — and the respondent is the one carrying the legal bill. The surge is real: the Commission's workload is up more than 70 per cent in three years, which the President attributes principally to AI making it a ten-minute job to file, and members reported signs of AI in 50 to 75 per cent of the cases they handled in the six months to March 2026. We covered how AI turned filing a claim into a ten-minute job. The Khan order is the reassuring half: the Commission will now make a careless applicant pay when they run a hopeless case on unchecked AI.

The other half is your exposure. Most small employers defend these claims themselves, or lean on a bookkeeper or HR adviser — often with ChatGPT open in the next tab. The standard the Commission now demands of applicants applies to your defence too. A fabricated citation doesn't just get struck out; AI invents authorities and misstates the law with total confidence, and once a decision-maker catches one in your paperwork, they doubt everything else you filed. If you're not sure how much of your business already runs on unverified AI output, see where your exposure sits.

What to do before 20 October

If AI touches anything you lodge with the Commission — a response, a submission, a witness statement — do three things, every time. Say you used it. Check every fact, section number, case name and quote against the primary source, not the chatbot that produced it. And make sure a witness statement is in the witness's own words. Then brief whoever helps you: a bookkeeper or HR consultant preparing documents on your behalf now carries the professional-representative standard, whether they realise it or not.

The wider lesson outlasts this one tribunal. A human checking AI output before it leaves the building is the cheapest insurance in the business — and it belongs on anything a customer, a court or a regulator will read, not just a Fair Work filing. If you'd rather build that habit deliberately than learn it in a hearing, a short conversation will show you where to start.