Skip to main content
Australian Workplace Strategies

Workplace Advisory & Compliance

AI in Fair Work disputes: lessons for employers from recent decisions

Four recent Commission matters show why employers should separate substance from volume, verify every filing and keep human judgment in control when AI enters a workplace dispute.

By the AWS Editorial Team
Two workplace professionals reviewing material on a laptop
Human review remains essential when AI-generated material is used in workplace disputes.

Key points

  • The Commission's recent decisions do not create a rule that using AI attracts costs or makes a dismissal valid; each outcome turned on the applicable statutory test and the conduct proved.
  • Khan v Aldi concerned persistence with an untenable minimum-employment-period claim after clear explanations and warnings; the costs ground was unreasonable continuation under s 400A, not AI use itself.
  • Anwar emphasises that generated material cannot replace the human judgment needed to identify the real issue, select relevant material and conduct litigation proportionately.
  • In De Bruyn, AI-appearing correspondence obscured reasonable concerns. A separate stop-bullying order required accurate and respectful communications, barred AI tools in preparing correspondence between those parties, and said the parties should endeavour to keep communication brief, to the point and meaningfully responsive.
  • Wibmer involved AI-assisted grievance communications and submissions, but the unfair dismissal outcome rested on the wider course of conduct assessed by the Commission, not on a standalone prohibition against AI.
  • From 20 October 2026 the Commission's final guidance will require disclosure, verification and, for witness material, authenticity confirmations when generative AI is used in documents lodged in Commission cases.

Four decisions, four different legal outcomes

Generative AI is now visible in workplace grievances, witness material and Fair Work Commission filings. Four decisions from 2026 show the practical problem, but they do not establish a single rule that AI use attracts costs, justifies dismissal or invalidates a document. The statutory question and the proved conduct remained decisive in each matter.
In Sadnan Khan v Aldi Pty Ltd [2026] FWC 3144, the Commission made a costs order after an applicant persisted with an untenable unfair dismissal claim. In Mr Zain Anwar v Rasier Pacific Pty Ltd [2026] FWCFB 142, a Full Bench addressed unreviewed generated material and the need for human judgment in conducting litigation. In Application by Mrs Deidre De Bruyn [2026] FWC 2887, the Commissioner found bullying, criticised correspondence that appeared to be AI-assisted and made a separate, party-specific stop-bullying order. In Mr Marcus Wibmer v Fujifilm Data Management Solutions Pty Ltd [2026] FWC 835, AI-assisted grievance communications and submissions formed part of a much wider factual record in an unsuccessful unfair dismissal application.
The common employer lesson is procedural rather than technological: identify the real issue, check facts and authorities, preserve authentic evidence and ensure a responsible person exercises judgment. Our separate guide to AI governance and human oversight in workplace decision-making addresses system selection, privacy and oversight. This briefing stays with complaints and Commission disputes.

Khan v Aldi: costs followed unreasonable persistence, not AI use alone

Sadnan Khan v Aldi Pty Ltd [2026] FWC 3144 was an unfair dismissal costs matter before Deputy President Easton. The applicant had not completed the statutory minimum employment period when he was notified of dismissal. Commission staff and the Deputy President explained the jurisdictional problem and warned of the possible costs consequences, but the claim continued on an argument directed to the wrong date. At the hearing, the applicant conceded that he had not completed the minimum period and discontinued the application. The published decision records that an appeal was lodged; no appeal disposition was identified at the time of publication.
The Commission then ordered the applicant to pay $1,230 towards Aldi's costs. Precision matters here. Section 400A permits a costs order where an unreasonable act or omission in connection with the conduct or continuation of a matter caused the other party to incur costs. The operative conduct was persisting after the claim's lack of prospects should have been apparent, despite explanations and warnings. Reliance on AI helped explain the irrelevant argument and failure to engage with the decisive issue; AI was not itself the statutory costs ground.
Employers should therefore avoid writing a costs threat around suspected AI use. Identify the dispositive issue, explain it accurately, give the other party a fair opportunity to reconsider, make any formal offer or warning proportionate, and obtain legal advice before seeking costs. A weak claim is not automatically unreasonable, and an unsuccessful party does not ordinarily pay costs in the Commission merely because they lose.

Anwar v Rasier Pacific: generated volume cannot replace judgment

Mr Zain Anwar v Rasier Pacific Pty Ltd [2026] FWCFB 142 arose from an unfair deactivation matter involving a digital labour platform worker, not an unfair dismissal claim. The Full Bench refused applications seeking correction or revocation of its earlier decision and addressed extensive material submitted without adequate revision.
Its central observation was that generative AI is a tool and is not capable of making the decisions required to conduct litigation effectively: human judgment is required. The practical concern was not polished prose. It was pages of generated material that obscured the actual complaints and required the Commission to identify the issues from the volume presented.
For employers, the same discipline applies on the responding side. A response should not answer every generated proposition at equal length. Build an issues schedule: allegation, material fact, employer position, evidence, legal relevance and next step. Address the claim the Commission can determine, while preserving any separate issues for the appropriate process. Our guide to responding to an unfair dismissal application sets out the broader response discipline.

De Bruyn: AI-appearing correspondence and a party-specific order

Application by Mrs Deidre De Bruyn [2026] FWC 2887 was an anti-bullying matter decided by Commissioner McKinnon. The Commission found that the applicant had been bullied at work and decided to make stop-bullying orders. In discussing the failed communication between the caretaker's representative and the body corporate, the Commissioner observed that correspondence appearing to have been prepared with AI was lengthy, wide-ranging, repetitive, generalised and often accusatory. The other party's disengagement was described as unsurprising, and the breakdown was detrimental to everyone involved.
Do not confuse that proceeding with the earlier Application by Mrs Deidre De Bruyn [2025] FWC 1011 (AB2024/725), which Commissioner Durham dismissed under s 587(1)(c) after finding that the parties had reached a binding settlement in November 2024; that decision contains no AI finding or prohibition.
Paragraph [27] of the reasons contains the AI-related observations: the Commissioner criticised correspondence that appeared to have been prepared with AI because it was lengthy, repetitive, generalised and accusatory, while recognising that reasonable underlying concerns should have been addressed. The separate stop-bullying order PR812696, made under s 789FF, then required all future communication between these parties to be accurate and respectful and barred the use of AI tools in preparing correspondence between them. It also said the parties should endeavour to keep their communications brief and to the point, and respond as necessary to issues raised in a meaningful way. The order took effect immediately.
That order was tailored to the parties and communication failure in this proceeding. It was not a general prohibition on AI-assisted workplace communication or on AI use in submissions to the Commission, and it was not a costs order; the $1,230 costs order belongs to Khan v Aldi and arose under a different provision on different facts. The employer lesson is equally balanced: dense correspondence should be controlled, not ignored. Acknowledge receipt, isolate and address the substantive complaints, ask targeted questions where necessary, and keep replies accurate, respectful, concise and responsive. If correspondence itself creates a psychosocial or conduct issue, manage that issue separately from the underlying complaint.

Wibmer v Fujifilm: AI-assisted grievances within a wider course of conduct

Mr Marcus Wibmer v Fujifilm Data Management Solutions Pty Ltd [2026] FWC 835 concerned a long-serving senior developer whom the employer dismissed for serious misconduct. The dispute began with a workplace disagreement and bullying complaint, then expanded through a large volume of communications and additional allegations. Mr Wibmer acknowledged relying on AI to prepare emails. Meeting attendees believed he was using AI to frame responses during meetings, and he did not challenge that evidence.
Deputy President Slevin dismissed the unfair dismissal application. The outcome did not turn on a rule that AI-assisted grievance emails warrant dismissal. The Commission found a valid reason in the wider course of conduct, while finding that the employer's failure to notify Mr Wibmer expressly of the decisive ‘ungovernable’ allegation and give him a direct opportunity to answer it weighed in favour of unfairness. Those procedural factors, his age and long service were ultimately outweighed by the misconduct. The reasons also criticised his AI-assisted communications and submissions, but that criticism was not a substitute for the statutory analysis. On 2 July 2026, a Full Bench refused permission to appeal in Wibmer v Fujifilm Data Management Solutions Pty Ltd [2026] FWCFB 159 after finding no arguable case of appealable error and no public interest basis to grant permission.
An employer should not label a worker difficult or unmanageable merely because a complaint is long, legalistic or AI-assisted. It should first test whether the conduct creates a separate, clearly articulated workplace issue, identify the evidence, put that issue to the employee and consider the response. Where dismissal is contemplated, use the record and fairness controls in our guide to fair and documented termination processes.

How to respond to a sprawling grievance

Start with extraction, not rebuttal. Read the whole communication once, then create a neutral schedule of the factual complaints: who is said to have done what, when, where, who may have witnessed it, what document may exist and what outcome is requested. Remove duplicated formulations without deleting distinct allegations. Send the schedule to the complainant and invite correction before scope is settled.
Triage each item into a response pathway. Some allegations may require investigation; some may be requests for information, review or adjustment; some may be employment-law assertions needing advice; some may be unsupported characterisations that do not require a finding. Explain that classification. The method in our early complaint triage guide and the reliability checks in reviewing workplace investigation findings remain applicable regardless of how the grievance was drafted.
Do not allow suspected AI use to become a credibility shortcut. Verify the underlying facts. Preserve the original message and attachments, the employer's issue schedule, clarification exchanges, scope decisions and responses. If a cited authority matters to the employer's position, open the primary decision or legislation and read the relevant passage; do not rely on the complainant's summary or on a generated employer summary.

Check the employer's own AI-assisted material

The Commission's concern applies to respondents as well as applicants. An employer response, chronology, witness statement or submission can be fluent and still contain the wrong date, invented authority, unsupported inference or a quotation detached from its context. Assign one accountable reviewer who understands both the facts and the proceeding, and require that person to check every material assertion against the source record.
Witness authenticity needs its own control. AI may help organise a chronology or improve expression, but a witness statement must remain the witness's evidence. Interview the witness in their own words, retain source notes, mark later changes, have the witness read the complete final statement and confirm that it is accurate and reflects their recollection. Never ask a tool to fill an evidentiary gap or make language more certain than the witness is.
Protect confidentiality. Do not paste employee health information, complaint material, privileged advice, commercial information or unredacted witness accounts into a public AI service without an approved use case and appropriate contractual, privacy and security controls. Keep a record of the tool, purpose, inputs at an appropriate level, reviewer, verification performed and final changes. That audit trail supports disclosure and explains how the filed document was produced without retaining unnecessary sensitive prompts.

The Commission's final guidance from 20 October 2026

On 24 August 2026 the Commission announced its final Guidance Note: Use of Generative Artificial Intelligence in Commission Cases. The requirements commence on 20 October 2026 and are not yet in force at this article's publication date of 24 September 2026. They apply in all Commission matters when a person uses generative AI to create, write, edit, modify or otherwise prepare an application or another document to be lodged. Using it only for spelling, grammar or formatting is excluded.
From commencement, the document must state that generative AI was used and explain how. A person — not the same or another AI tool — must check that facts and evidence exist and are correct, that legal sources exist and support the stated propositions, and that extracts are accurate; the document must state that the checks were completed. For a witness statement or declaration, the witness or declarant must also state that it is based on their own knowledge and reflects their own words, and declare that it is true to the best of their knowledge. A professional representative, expressly including an HR adviser employed by the represented party, must hyperlink all referred case law where links are available.
The Guidance Note also says personal information about another person and confidential case information should not be given to public or insecure generative AI. Non-compliance may affect the case: a document may receive less weight or be disregarded, costs may be ordered, or the case may be dismissed. These are possible case-management or legal consequences, not automatic penalties for using AI. The Commission's official generative AI guidance page should be checked when material is prepared because the Guidance Note may be amended. The Commission said it would modify forms and submission templates in individual rights matters to add an AI disclosure and declaration, effective from 20 October 2026. Employers should adopt the verification disciplines now and plan for disclosure in documents lodged from that date.

Sources, limits and how AWS can help

This briefing uses the Commission's published reasons for Sadnan Khan v Aldi Pty Ltd [2026] FWC 3144 (Deputy President Easton, 19 August 2026), Mr Zain Anwar v Rasier Pacific Pty Ltd [2026] FWCFB 142 (Full Bench, 19 June 2026), Application by Mrs Deidre De Bruyn [2026] FWC 2887 (Commissioner McKinnon, 30 July 2026), the separately published De Bruyn stop-bullying order PR812696, Mr Marcus Wibmer v Fujifilm Data Management Solutions Pty Ltd [2026] FWC 835 (Deputy President Slevin, 18 March 2026), and the Wibmer appeal decision [2026] FWCFB 159 (2 July 2026), together with the final Guidance Note and the Commission President's 24 August 2026 statement. The briefing distinguishes the reasons from the operative order, and does not treat their observations or party-specific terms as a separate legal test or transplant the outcome of one case into another.
AWS supports employers through workplace investigations and workplace advisory and compliance: triaging complex grievances, defining investigation scope, reviewing the evidence and decision record, and building filing protocols that require human verification. AWS is a workplace consultancy, not a law firm. This article is general information current at 24 September 2026; obtain legal advice on the jurisdiction, merits, evidence and strategy in a particular Commission matter.

Employer protocol for AI-affected disputes

  • Preserve the original grievance, attachments and metadata before transforming it into a summary or issues schedule.
  • Extract each distinct factual complaint and requested outcome into a neutral issues schedule, remove repetition without discarding substance, and ask the complainant to confirm or correct it before finalising investigation scope.
  • Separate investigation issues, legal assertions, information requests, adjustment requests and conduct concerns into the right response pathways.
  • Nominate one contact, a proportionate communication process and a realistic timetable for updates and responses.
  • Open and read every statute, case and policy that materially supports the employer's position; never verify a citation from a generated summary alone.
  • Check every date, quotation, amount and factual proposition in a proposed filing against the source record.
  • Interview witnesses in their own words, retain source notes and require them to approve the complete final statement.
  • Do not upload confidential, privileged or sensitive employee information to an unapproved public AI tool.
  • Record the tool used, purpose, responsible reviewer, verification completed and material changes made to any AI-assisted filing.
  • Apply the Commission's disclosure, verification, authenticity and hyperlinking requirements to documents lodged from 20 October 2026.
  • Keep suspected AI use separate from the legal merits, disciplinary question and any application for costs.
  • Have a person with authority and subject-matter competence exercise final judgment before correspondence, evidence or submissions are issued.

Frequently asked questions

Does using AI in a Fair Work Commission case automatically expose a party to costs?
No. Khan v Aldi applied the existing costs power in s 400A to unreasonable conduct that caused the other party to incur costs: continuing a claim that could not succeed after the eligibility problem and possible costs consequences had been explained. AI use was part of the factual context, not a new or automatic ground for costs.
Can an employer ignore a grievance because it appears to have been written by AI?
No. The employer should identify and respond to the substantive allegations, requests and safety issues, even where the document is repetitive or legally unfocused. A short issues schedule, clarification questions and an agreed response process can control volume without dismissing a potentially valid complaint because of how it was drafted.
What changes at the Fair Work Commission on 20 October 2026?
The Commission's final guidance will apply to anyone using generative AI to create, write, edit or otherwise prepare a document lodged in a Commission case. It requires disclosure of the use and how the tool was used, personal verification of accuracy, relevance, evidence and authorities, and additional authenticity confirmations for witness statements or declarations. Professional representatives must also hyperlink cited cases.
Can HR use AI to draft an employer response or witness statement?
AI may assist with drafting, but the employer and any representative remain responsible for the document. Facts must be checked against the record, authorities must be opened and read, confidential information must be protected, and a witness must confirm that their statement reflects their own knowledge and words. From 20 October 2026, Commission filings must also meet the final guidance requirements.

Discuss this matter with AWS

Briefings can be scoped on a confidential basis. We respond within two business days.

Contact AWS

Related briefings

Workplace Advisory & Compliance

How to conduct a wage compliance review

An end-to-end guide for employers on governing, scoping, calculating, verifying and acting on a wage compliance review.

Read briefing →

Workplace Advisory & Compliance

How to determine award coverage and classification

Coverage, application and classification are three separate questions, and job titles, salaries and payroll codes answer none of them. A step-by-step method for reaching a position you can stand behind.

Read briefing →

Workplace Advisory & Compliance

Preparing for enterprise bargaining: an employer's roadmap

Making a single-enterprise agreement is a statutory process with fixed steps and deadlines. This roadmap follows the pathway from scoping and notification time through to approval and payroll implementation.

Read briefing →