The Fair Work Ombudsman and the Fair Work Commission have both published official guidance on using AI for workplace questions and Commission cases. The Commission has gone as far as publishing copy-and-paste prompts to help parties get better answers out of ChatGPT. This is not a ban. It’s an instruction manual. Here’s what it says, how employers can use it during a claim, and the risks the guidance itself warns about. 

The regulator has picked a lane 

Faced with a flood of AI-prepared claims, Fair Work had two options: fight the tide or channel it. It chose to channel it. 

The Fair Work Ombudsman now has a dedicated page on using AI for help with workplace questions. The Fair Work Commission has gone further, publishing a full Guidance Note, a video, verification checklists and, remarkably, a set of set-up prompts it wants parties to paste into their GenAI chats before asking about a case. The Commission was explicit that this is a deliberate choice. After 28 consultation submissions, it dropped the blanket cautionary framing and built materials that emphasise responsible and effective use instead. 

Why does that matter to you as an employer? Because it means AI-prepared claims are now officially sanctioned, structurally supported, and about to get better. The applicant sitting across from you at conciliation didn’t just use ChatGPT. They used ChatGPT configured with prompts written by the tribunal deciding your case. You need to know what the guidance says, because both sides of every claim are now playing by it. 

AI has already arrived in the Commission 

This isn’t a future problem. It’s a current one. 

The Fair Work Commission’s own research, released alongside its new Guidance Note in August 2026, found that roughly 40 per cent of surveyed applicants used generative AI to prepare or manage their case. ChatGPT was the most common tool, used by about 77 per cent of those AI users, and around 60 per cent were relying on the free tier. 

Commission Members reported spotting signs of AI use by one or both parties in 50 to 75 per cent of the cases they handled in the six months to March 2026. 

President Justice Adam Hatcher has attributed the Commission’s “unprecedented growth” in workload principally to the spread of GenAI tools among applicants, not to any shift in the labour market. By the end of the 2025–26 financial year, the Commission’s total caseload is expected to have grown by more than 70 per cent over three years. 

Read that again. AI hasn’t just changed how claims are prepared. It’s changed how many claims get lodged. Employees who would never have drafted an unfair dismissal application themselves are now doing it in an evening with a chatbot. Some of those claims have merit. Many don’t. The Commission’s research found that AI-dependent applicants are more likely to lodge claims with weaker merit, inflated settlement expectations and, in some cases, fabricated case citations. 

For employers, that means two things: 

• You’re more likely than ever to be on the receiving end of a claim. 

• When you respond, you’ll probably be tempted to use the same tools. And now there are formal rules about how you do that. 

The new rules: what applies from 20 October 2026 

The Commission’s Guidance Note on the use of generative AI applies to all Commission matters from 20 October 2026, including unfair dismissal and general protections applications. It covers ChatGPT, Claude, Copilot, Gemini and any similar tool. 

The rules apply equally to applicants and respondents. If your business uses GenAI to create, write, edit or otherwise prepare an application, submission, witness statement, employer response or any other document in a case, you must: 

1. Disclose. Tell the Commission that GenAI was used and explain how. Some Commission forms will have a specific question about it. For documents without the question, you add a note at the beginning, along these lines: “GenAI was used in preparing this document. I used it by asking GenAI to [insert details].” 

2. Verify. Check the document yourself to make sure all information is accurate, relevant, and supported by real evidence and real legal authorities. Then state in the document that you’ve completed these checks. You can’t outsource this. The Commission is explicit that you remain responsible for everything you lodge. 

3. Keep witness statements authentic. For witness statements and declarations, the witness must confirm the document is based on their own knowledge, reflects their own words, and is true to the best of their knowledge. An AI-polished statement that no longer sounds like the witness is a problem, and it will get pulled apart in cross-examination. 

There’s one more obligation that should make every HR team sit up. Legal practitioners and other professional representatives, including in-house HR advisers representing a party, must include hyperlinks to any case law referred to in documents prepared using GenAI. If you’re the HR manager running the employer’s response, that’s you. 

The consequences of getting it wrong are not theoretical: 

• Documents may be given less weight or disregarded entirely 

• The Commission can order you to pay the other party’s costs as well as your own 

• Your case can be dismissed 

• Knowingly giving false or misleading information to the Commission is a criminal offence under section 137.1 of the Criminal Code, punishable by up to 12 months’ imprisonment 

Costs orders in this space are already happening. A recent decision saw a worker ordered to pay ALDI’s costs after running a dismissal claim on AI. The scrutiny cuts both ways. 

How employers can legitimately use AI during a claim 

None of this means AI is banned. The Commission has deliberately chosen not to frame the reform as a crackdown. The final materials emphasise responsible and effective use rather than discouraging it. Used carefully, AI can genuinely help a business respond to a claim faster and more coherently. 

Here’s where it earns its keep. 

1. Understanding the process 

If you’ve never faced an unfair dismissal application, the process itself is disorienting. Forms, deadlines, conciliation, directions, hearings. AI is a decent starting point for plain-English explanations of how the Commission works, what a Form F3 employer response is, what happens at conciliation, and what a jurisdictional objection means. Treat it as orientation, not advice. 

2. Organising facts and building a chronology 

This is arguably the safest and most valuable use. Feed AI a de-identified timeline of events (more on de-identification below) and ask it to organise the facts into a clear chronology, flag gaps, and identify which events need supporting documents. Claims are won and lost on documentation. AI is good at helping you see where your paper trail is thin. 

3. Drafting and editing 

AI can produce a first draft of a submission or tidy up your own rough draft. It can restructure a rambling account into something a Commission Member can actually follow. It can translate, transcribe, and edit. All of these are uses the Commission itself lists as legitimate, provided they’re disclosed. 

4. Stress-testing your position 

The Commission’s research draws a line between disciplined “GenAI-assisted” users, who cross-check outputs and prompt the tool to argue against their own case, and “GenAI-dependent” users, who accept fluent, confident answers at face value. Be the first kind. Ask the tool to make the applicant’s best argument. Ask it what weakens your position. That’s where AI adds real analytical value, and it’s the exact opposite of how most people use it. 

5. Use the Commission’s own set-up prompts 

The FWC has published a set of set-up prompts to paste into a GenAI chat before asking about a case. They’re designed to force the tool toward accuracy, Australian sources and honesty about weak positions. Here they are in full, exactly as published at fwc.gov.au/genai. Copy everything below and paste it into the chat before you start: 

“Apply the following rules to every response: 

• Base responses on the Fair Work Act 2009, information published by the Fair Work Commission (fwc.gov.au), the Fair Work Ombudsman (fairwork.gov.au), and relevant Australian court and tribunal decisions. 

• Prioritise accuracy and objectivity over telling me what you think I want to hear. 

• Do not invent laws, cases, authorities, facts or events. If authorities conflict, explain the conflict. If available authorities do not support my position, say so clearly. 

• Treat my statements as unverified information, not established facts. You may draw reasonable inferences but clearly identify them as inferences. 

• Do not assume my case is strong or weak based on how I describe it. 

• Where relevant facts are missing, explain: what is known, what is unknown, what additional information is needed, and why it matters. 

• Ask questions where further information is required instead of making assumptions. 

• Clearly distinguish between: facts provided, legal information, procedural information, and analysis or assessment. 

• Explain any uncertainty in the law, the evidence or the available information, and explain how it affects the answer. 

• For each legal source cited, provide its name and explain how it can be verified. Where possible, include links to official sources such as AUSTLII or court and tribunal websites. 

• Use plain language and be concise. Avoid emotional validation, adversarial framing and unnecessary legal jargon. 

• If I ask about making an application or claim, first explain which body, if any, has authority to deal with the case. Then establish eligibility to apply before assessing the merits. 

• If my position is not supported by the facts or available legal authorities, tell me that clearly and explain why. 

• Do not predict case outcomes with certainty. Where relevant, explain factors that may strengthen or weaken the case and any limitations in the available information.” 

Look at what these prompts are actually doing. Almost every line is a defence against a known failure mode. “Do not invent laws, cases, authorities, facts or events” exists because AI invents them. “Prioritise accuracy over telling me what you think I want to hear” exists because these tools flatter. “Do not assume my case is strong based on how I describe it” exists because they take your version of events as gospel. The tribunal has effectively published a list of everything that goes wrong when people use AI unprompted. Treat the prompt list as a risk register. 

Two caveats from the Commission itself. First, even with the prompts, output can still be incomplete or wrong. Second, GenAI tools forget instructions in long conversations, so if answers start drifting, re-paste the prompts or start a new chat. 

Now the caveats. All of them. 

This is the part of the article that matters most, so we’re not going to soften it. AI in a Fair Work claim is a loaded tool, and here is every way it goes wrong. 

It makes up cases 

GenAI fabricates legal authorities. Confidently, fluently, with plausible-looking citation formats. The Commission has already dealt with an unfair dismissal matter where an applicant’s AI-generated citations turned out not to exist. The Commission’s guidance tells parties to verify every case on AustLII or the Commission’s own document search, and every piece of legislation on the Federal Register of Legislation. If you can’t find it, it may not be real. Lodging a submission built on fake authority is how documents get disregarded, how costs orders happen, and how you end up explaining yourself under section 137.1. 

It doesn’t know Australian law reliably 

The Fair Work Ombudsman’s guidance on this is blunt: AI-generated workplace information can be incorrect, incomplete, outdated, or simply not about Australia. These tools blend material from different countries, industries and time periods and present it as fact. US at-will employment concepts bleed into answers about Australian dismissal law. State and federal rules get mixed. None of it comes with a warning label. 

It doesn’t know what the law is today 

Awards change. Rates change. Legislation gets amended, and sometimes AI describes changes that haven’t commenced yet as if they’re already law. The FWO gives the example of an employee, Sam, who used a Google AI summary and concluded he was being underpaid. He wasn’t. The AI had served up information about upcoming award changes, not his current entitlements. Now imagine that error running the other direction, inside your employer response. 

It tells you what you want to hear 

This one is subtle and dangerous. Large language models are agreeable by design. Describe your dismissal decision in your own words and the tool will usually tell you it sounds reasonable. The Commission’s research found that AI-dependent users develop inflated expectations precisely because the tool validates them. Employers are not immune. An AI that assures you your process was procedurally fair is not a substitute for someone who will actually tell you it wasn’t. 

Confidentiality and privacy 

The Commission’s warning is direct: do not enter personal information about yourself or other people, or confidential case information, into public GenAI tools. Whatever you type may not remain private or secure. For an employer, the stakes are higher than for an individual. You hold the applicant’s personal information, other employees’ details, medical information, payroll data. Pasting a termination letter or an investigation report into a free chatbot is a privacy problem on top of a litigation problem. If you use AI at all, de-identify everything first, and think hard about which tool and which tier you’re using. 

Witness statements that aren’t the witness 

If AI rewrites a witness statement into polished prose the witness would never say, you’ve created a document the witness can’t authentically own. Under the new rules the witness must confirm it reflects their own knowledge and their own words. A statement that collapses at hearing because the witness doesn’t talk like that, or can’t explain what a paragraph means, damages your whole case. 

Disclosure isn’t optional, and non-disclosure is discoverable 

Commission Members are already identifying AI use in the majority of matters they handle. Assume they can tell. The research found people disclose more willingly when they believe it won’t be held against them, and the Commission has designed the regime accordingly. Honest disclosure of sensible AI use is fine. Getting caught having not disclosed it is a credibility problem you don’t recover from. 

The verification burden lands on you 

Here’s the uncomfortable truth about the new rules. The verification requirement assumes you can actually verify. Checking whether a cited case is real is the easy part. Checking whether it genuinely supports the proposition it’s cited for requires reading the decision and understanding it. Checking whether an AI-drafted jurisdictional argument is right requires knowing the law. Most HR teams, and virtually all small business owners, can’t do that check meaningfully. Which means AI gets you a document that looks finished but that you cannot honestly certify. That’s the trap. 

AI is a starting point, not advice 

Both regulators land in the same place. The Fair Work Ombudsman says use AI as a starting point only, don’t rely on it as a final source, and go to trusted sources for the answer. The Commission says GenAI should not be used as a substitute for legal advice. When the national workplace regulator and the national workplace tribunal both attach the same disclaimer, believe them. 

The bottom line for employers 

Used well, AI can help you get organised, understand the process, and draft faster. Used badly, it can hand the Commission a reason to disregard your documents, order costs against you, or dismiss your case, while you certify checks you weren’t equipped to perform. 

The dividing line is simple. AI can help you prepare. It cannot tell you whether your dismissal was defensible, whether your process was procedurally fair, whether to fight or settle, or what a claim is actually worth. Those are judgment calls, and they need a human who knows Australian workplace law and will give it to you straight. 

Talk to a human before you lodge anything 

That’s exactly what the HR Command Workplace Advice Line is for. 

When a claim lands, or when you’re about to make a decision that could trigger one, you get direct access to qualified workplace relations advice. Real answers about your actual situation. Whether your process stacks up. What your response should say. What to verify before anything goes to the Commission. No hallucinated cases, no US law, no confident nonsense, and nothing you have to disclaim in a footnote. 

Pair that with HR Command’s compliant templates, policies and documentation, and you’re not scrambling to build a paper trail after the application arrives. You already have one. 

Use AI for the legwork. Use us for the advice. 

Book a demo at hrcommand.com.au/demo and see how the Workplace Advice Line and the full HR Command platform keep you covered before, during and after a Fair Work claim. 

This article is general information, not legal advice. Every claim turns on its own facts. If you’re facing a Fair Work matter, get advice on your specific situation.