An Australian Tribunal Will Soon Make You Declare If You Used AI

Australia’s workplace tribunal has done something most institutions have only talked about: it has put a price on bad AI advice. Sadnan Khan, a former ALDI employee challenging his dismissal, was ordered to pay 1,230 Australian dollars towards the supermarket’s legal costs after the Fair Work Commission found he had used AI as a “quasi legal advisor” and pressed on despite repeated warnings that his case had no substantial prospects of success. Deputy President Michael Easton called the step rare, and added a line that will be quoted for a while: “If Mr Khan had properly read his own AI generated replies, he would have known that his case was doomed.”

The case landed alongside a report the tribunal commissioned into how litigants are actually using these tools, and the numbers explain the urgency. Generative AI was part of the reason its caseload grew by 40% between 2023-24 and 2024-25. Around 40% of surveyed cases involved a litigant using AI. More than three quarters of that group used ChatGPT, and 60% of those used the free version. People who do not speak English as a first language were twice as likely to use AI as native speakers. Khan, originally from Bangladesh, told the ABC he had used a paid ChatGPT to make sense of the case law being quoted at him, and admitted he forgot to strip chat instructions out of his submissions. From 20 October 2026, applicants will have to disclose any AI use, verify facts, authorities and hyperlinks, and confirm witness evidence, with consequences flagged for those who do not.

What’s actually going on here: the tribunal is being careful not to say AI is the problem, and that nuance matters. Its own decision notes that AI is helping people with genuine claims who might otherwise have been put off entirely, and that sophisticated users treat it as one tool among many. The contrast in the ABC’s reporting is stark. Macquarie University lecturer Gregory Baker used a team of AI agents to become the first person to successfully challenge Australia’s casual employment laws, and he did it by treating his filings like a software project with a checking step for every citation. Asking a chatbot as an oracle, he said, does “a terrible job.” Monash law professor Genevieve Grant points at the mechanism behind the failures: chatbots have a sycophantic tendency, meaning they lean towards agreeing with you, which can convince someone their claim has legs when it does not.

What this means for you: you do not need to be in a tribunal for this to apply. Whenever you use a chatbot for something with a real consequence, a contract, a complaint, a tax question, a medical form, assume it will tell you what you want to hear unless you make it argue the other side. Ask it what the strongest case against you is. Check every citation it gives you, because a confident reference to a case or a rule is exactly the kind of thing these models invent. And read what it wrote before you send it, all of it. Both cautionary details in this story, the unremoved chat instructions and the unread replies, came down to that one step.

Sources

• Fair Work Commission condemns ‘plain wrong’ AI legal advice as cases with AI litigants surge (ABC News)
• Fair Work Commission sets new rules for AI use in cases (Human Resources Director)
• Fair Work Commission cracks down on dodgy generative AI cases (SmartCompany)