The Fair Work Commission condemns 'plain wrong' AI legal advice after a sacked Aldi worker used ChatGPT as a "quasi-legal advisor" to fight his dismissal and lost, then got ordered to help pay his former employer's legal bill. The ruling lands as the Commission rolls out a mandatory AI-disclosure rule for every unfair dismissal and general protections case filed from October 20, 2026.
Short answer: In August 2026, Fair Work Commission Deputy President Michael Easton ruled that AI-generated legal arguments used by former Aldi worker Sadnan Khan were "just plain wrong," ordering him to pay $1,230 toward Aldi's costs. The case coincides with a roughly 40% surge in Commission filings and a new rule, effective October 20, 2026, requiring applicants to disclose any AI use and verify what it produced.
What happened in the Khan v Aldi case
Sadnan Khan was dismissed from his job at Aldi and took the case to the Fair Work Commission, arguing he'd been unfairly let go. According to ABC News, Khan leaned on ChatGPT throughout the process, effectively using it as his lawyer rather than as a drafting aid — asking it to build his legal argument, not just tidy up his wording.
Deputy President Michael Easton wasn't gentle about the result. He found the AI-generated case had "no substantial prospects of success" and was, in his words, "just plain wrong." Easton noted that "if Mr Khan had properly read his own AI generated replies … he would have known that his case was doomed." Khan was ordered to pay $1,230 toward Aldi's legal costs — a cost order against an unsuccessful applicant is unusual in the Commission and signals how seriously Easton treated the misuse. Khan later reflected that "the main thing AI suffers is they do things not the Aussie [court] way," which is a fair summary of the core problem: a chatbot trained mostly on generic legal reasoning doesn't know Australian employment law procedure, evidentiary rules, or the Commission's own precedent.
Why this is happening now: AI use in Commission cases is surging
This isn't an isolated incident. The Commission has seen roughly a 40% jump in case filings between the 2023-24 and 2024-25 periods, and AI is a meaningful driver of that spike, according to the same ABC reporting. In a sample of surveyed cases, around 40% of litigants had used AI somewhere in preparing their case. Of those, more than 75% relied on ChatGPT specifically, and about 60% of ChatGPT users were on the free tier — the version most prone to confident, unverified answers with no live legal database behind it. Non-English speakers were roughly twice as likely as native English speakers to lean on AI tools, which points to something more sympathetic than laziness: people without ready access to a lawyer are reaching for the cheapest available help, and getting a plausible-sounding argument that doesn't hold up.
Monash University law professor Genevieve Grant has flagged the same pattern researchers see across other courts and tribunals worldwide: self-represented litigants use AI because free legal advice is scarce, and the tools rarely flag their own uncertainty.
Step-by-step: what changes under the FWC's new AI disclosure rule
The Commission published a final Guidance Note on generative AI use following a consultation period that drew industry submissions after the draft was released in March 2026. Per SmartCompany’s coverage, here's what actually changes, in order, from October 20, 2026:
1. You must disclose AI use in the document itself
Any application, submission, or witness statement prepared with the help of ChatGPT, Claude, Copilot, Gemini, or similar tools needs a disclosure statement saying so. Using AI purely for spelling or grammar checks is exempted — the rule targets AI that shapes the substance of an argument, not AI that proofreads it.
2. You must verify every fact, case reference, and hyperlink
The guidance specifically calls out "hallucinated" case citations — AI tools inventing plausible-sounding but nonexistent case law is a well-documented failure mode, and the Commission wants applicants to state, in writing, that they checked.
3. Witness evidence has to match what the witness actually knows
If AI helped draft a witness statement, the person signing it still has to confirm the content reflects their own firsthand knowledge, not language the AI generated on their behalf.
4. Keep confidential information out of public AI tools
The guidance separately warns against feeding personal or confidential case details into consumer AI products that don't guarantee how that data is stored or used.
Non-compliance isn't just a formality issue. Documents that fail these checks can be given reduced weight or disregarded entirely, costs can shift to the non-compliant party — exactly what happened to Khan — and, in the most serious cases, submitting false or misleading evidence can carry penalties under existing law.
I tested what ChatGPT actually gives you for an unfair dismissal question
In my testing, I asked ChatGPT's free tier a version of the question Khan was likely working from — "was I unfairly dismissed if my employer didn't give a written warning first?" The answer came back fast, confident, and structured like a legal memo, complete with numbered "grounds" for a claim. What it did not do, by default, was tell me whether those grounds matched current Fair Work Act thresholds, cite a real Commission decision, or flag that "no written warning" alone rarely wins an unfair dismissal case without other factors. When I pushed it to cite a specific precedent, it produced a case name that I could not verify existed. That's the exact failure mode Easton called out in Khan's case: fluent, organized, and wrong in ways a first-time reader has no way to catch.
Example prompts you can copy — for research, not for arguments
If you're using AI to prepare for a Fair Work Commission matter, use it to understand the process, not to write your legal argument for you:
- "Explain in plain English what 'harsh, unjust, or unreasonable' means under Australia's unfair dismissal law, and tell me you can't verify current case law without a citation check."
- "What documents does the Fair Work Commission typically expect in an unfair dismissal application, based on its public forms and guides?"
- "List the questions I should ask a free legal service (like a community legal centre) before my Fair Work Commission hearing."
- "Summarize the Fair Work Commission's own Guidance Note on generative AI use so I know what disclosure I need to make."
Always follow up with the Commission's own published guides and, where possible, a real duty lawyer or community legal centre — AI is a starting point for orientation, not a substitute for verified legal authority.
Common mistakes people make when leaning on AI for legal arguments
The single biggest mistake in Khan's case, per Easton's own ruling, was not reading the AI's output critically before filing it — the warning signs that the case was weak were reportedly visible in ChatGPT's own replies. The second is treating a chatbot's confident tone as a proxy for accuracy; free-tier models in particular will produce a complete-sounding argument with zero indication of how uncertain the underlying claim is. Third, people skip verification of case citations entirely, which is precisely the "hallucination" problem the Commission's new rule targets. Fourth, non-English speakers using AI as their primary translator and legal advisor are, per the Commission's own data, twice as exposed to this risk, which argues for pairing AI use with a free legal service rather than relying on it alone. Fifth — and this will matter a lot after October 20 — filing an AI-assisted document without disclosing it is now a compliance failure on its own, separate from whether the argument itself was any good.
Not every AI-assisted case fails: the Macquarie University counterexample
The Commission's own data shows AI isn't uniformly bad for self-represented litigants — it depends heavily on how it's used. A Macquarie University computer science lecturer, Gregory Baker, successfully used AI tools to help challenge the university's treatment of his casual employment status, a case working through appeal as of late August 2026. The difference wasn't the tool; it was that Baker, with technical background, appears to have used AI to organize and stress-test an argument he understood, rather than outsourcing the argument itself.
| Khan v Aldi | Baker v Macquarie University | |
|---|---|---|
| AI's role | Built the core legal argument | Assisted with research and organization |
| Outcome | Case failed; $1,230 costs order | Successful challenge; on appeal |
| Applicant's review of AI output | Reportedly did not catch AI's own warning signs | Applied domain understanding to verify claims |
| Lesson | AI as advisor, unchecked, produced a "plain wrong" case | AI as a research aid, checked by a knowledgeable user, held up |
Tools that make this easier — and safer — than what Khan did
If you're going to use AI anywhere near a legal or employment matter, understand exactly what the free tier is and isn't good for. Our guide to using ChatGPT covers the basics of what the tool can reliably do versus where it tends to overreach, and our roundup of free AI tools is a useful gut-check on which tools are genuinely free versus rate-limited demos. If you want the bigger picture on why courts and tribunals are tightening these rules, we've also covered a related case where someone hid a prompt injection inside a legal filing to try to manipulate an AI reviewer, and our piece on AI regulation and messaging tracks how institutions are responding to exactly this kind of misuse. For the labor-market backdrop driving more people toward self-representation and free AI tools in the first place, see our coverage of AI’s effect on entry-level jobs and why an AI jobs apocalypse probably isn’t imminent — and check our AI tool ratings before trusting any tool with something this consequential.
Frequently Asked Questions
Is it against the rules to use AI for a Fair Work Commission case?
No. AI use itself isn't banned. From October 20, 2026, you must disclose that you used it and confirm you verified the facts, case references, and any witness content it helped produce.
What actually happened in the Khan v Aldi case?
Sadnan Khan used ChatGPT to build his unfair dismissal argument against Aldi. Deputy President Michael Easton found the AI-generated case "just plain wrong" with no real prospects of success and ordered Khan to pay $1,230 toward Aldi's legal costs.
When does the Fair Work Commission's AI disclosure rule take effect?
October 20, 2026. It applies to individual rights matters, including unfair dismissal and general protections applications, and follows a consultation period after the draft guidance was released in March 2026.
What's the easiest way to avoid Khan's mistake?
Use AI to understand the process and organize research, not to generate your legal argument outright. Read everything it produces critically, verify every case citation independently, and disclose the AI use in your filing once the October rule takes effect.
Does using the free version of ChatGPT make this worse?
The data suggests it correlates with risk: about 60% of AI-using litigants on ChatGPT were on the free tier, which lacks the verification and up-to-date case-law access that would catch an invented citation before it's filed.